|
- Chunking in HDF5
+ @ref hdf5_chunking
|
Structuring the use of chunking and tuning it for performance.
diff --git a/doxygen/dox/chunking_in_hdf5.dox b/doxygen/dox/chunking_in_hdf5.dox
new file mode 100644
index 00000000000..b46662a3a5b
--- /dev/null
+++ b/doxygen/dox/chunking_in_hdf5.dox
@@ -0,0 +1,398 @@
+/** \page hdf5_chunking Chunking in HDF5
+ *
+ * \section sec_hdf5_chunking_intro Introduction
+ * Datasets in HDF5 not only provide a convenient, structured, and self-describing way to store data,
+ * but are also designed to do so with good performance. In order to maximize performance, the HDF5
+ * library provides ways to specify how the data is stored on disk, how it is accessed, and how it should be held in memory.
+ *
+ * \section sec_hdf5_chunking_def What are Chunks?
+ * Datasets in HDF5 can represent arrays with any number of dimensions (up to 32). However, in the file this dataset
+ * must be stored as part of the 1-dimensional stream of data that is the low-level file. The way in which the multidimensional
+ * dataset is mapped to the serial file is called the layout. The most obvious way to accomplish this is to simply flatten the
+ * dataset in a way similar to how arrays are stored in memory, serializing the entire dataset into a monolithic block on disk,
+ * which maps directly to a memory buffer the size of the dataset. This is called a contiguous layout.
+ *
+ * An alternative to the contiguous layout is the chunked layout. Whereas contiguous datasets are stored in a single block in
+ * the file, chunked datasets are split into multiple chunks which are all stored separately in the file. The chunks can be
+ * stored in any order and any position within the HDF5 file. Chunks can then be read and written individually, improving
+ * performance when operating on a subset of the dataset.
+ *
+ * The API functions used to read and write chunked datasets are exactly the same functions used to read and write contiguous
+ * datasets. The only difference is a single call to set up the layout on a property list before the dataset is created. In this
+ * way, a program can switch between using chunked and contiguous datasets by simply altering that call. Example 1, below, creates
+ * a dataset with a size of 12x12 and a chunk size of 4x4. The example could be changed to create a contiguous dataset instead by
+ * simply commenting out the call to #H5Pset_chunk and changing dcpl_id in the #H5Dcreate call to #H5P_DEFAULT.
+ *
+ * Example 1: Creating a chunked dataset
+ * \code
+ * #include "hdf5.h"
+ * #define FILENAME "file.h5"
+ * #define DATASET "dataset"
+ *
+ * int main() {
+ *
+ * hid_t file_id, dset_id, space_id, dcpl_id;
+ * hsize_t chunk_dims[2] = {4, 4};
+ * hsize_t dset_dims[2] = {12, 12};
+ * herr_t status;
+ * int i, j;
+ * int buffer[12][12];
+ *
+ * // Create the file
+ * file_id = H5Fcreate(FILENAME, H5F_ACC_TRUNC, H5P_DEFAULT, H5P_DEFAULT);
+ *
+ * // Create a dataset creation property list and set it to use chunking
+ * dcpl_id = H5Pcreate(H5P_DATASET_CREATE);
+ * status = H5Pset_chunk(dcpl_id, 2, chunk_dims);
+ *
+ * // Create the dataspace and the chunked dataset
+ * space_id = H5Screate_simple(2, dset_dims, NULL);
+ * dset_id = H5Dcreate(file_id, DATASET, H5T_STD_I32BE, space_id, H5P_DEFAULT, dcpl_id, H5P_DEFAULT);
+ *
+ * // Initialize dataset
+ * for (i = 0; i < 12; i++)
+ * for (j = 0; j < 12; j++)
+ * buffer[i][j] = i + j + 1;
+ *
+ * // Write to the dataset
+ * status = H5Dwrite(dset_id, H5T_NATIVE_INT, H5S_ALL, H5S_ALL, H5P_DEFAULT, buffer);
+ *
+ * // Close
+ * status = H5Dclose(dset_id);
+ * status = H5Sclose(space_id);
+ * status = H5Pclose(dcpl_id);
+ * status = H5Fclose(file_id);
+ * }
+ * \endcode
+ *
+ * The chunks of a chunked dataset are split along logical boundaries in the dataset's representation as an array, not
+ * along boundaries in the serialized form. Suppose a dataset has a chunk size of 2x2. In this case, the first chunk
+ * would go from (0,0) to (2,2), the second from (0,2) to (2,4), and so on. By selecting the chunk size carefully, it is
+ * possible to fine tune I/O to maximize performance for any access pattern. Chunking is also required to use advanced
+ * features such as compression and dataset resizing.
+ *
+ *
+ *
+ * |
+ * \image html chunking1and2.png
+ * |
+ *
+ *
+ *
+ * \section sec_hdf5_chunking_data Data Storage Order
+ * To understand the effects of chunking on I/O performance it is necessary to understand the order in which data is
+ * actually stored on disk. When using the C interface, data elements are stored in "row-major" order, meaning that,
+ * for a 2- dimensional dataset, rows of data are stored in-order on the disk. This is equivalent to the storage order
+ * of C arrays in memory.
+ *
+ * Suppose we have a 10x10 contiguous dataset B. The first element stored on disk is B[0][0], the second B[0][1],
+ * the eleventh B[1][0], and so on. If we want to read the elements from B[2][3] to B[2][7], we have to read the
+ * elements in the 24th, 25th, 26th, 27th, and 28th positions. Since all of these positions are contiguous, or next
+ * to each other, this can be done in a single read operation: read 5 elements starting at the 24th position. This
+ * operation is illustrated in figure 3: the pink cells represent elements to be read and the solid line represents
+ * a read operation. Now suppose we want to read the elements in the column from B[3][2] to B[7][2]. In this case we
+ * must read the elements in the 33rd, 43rd, 53rd, 63rd, and 73rd positions. Since these positions are not contiguous,
+ * this must be done in 5 separate read operations. This operation is illustrated in figure 4: the solid lines again
+ * represent read operations, and the dotted lines represent seek operations. An alternative would be to perform a single
+ * large read operation, in this case 41 elements starting at the 33rd position. This is called a sieve buffer and is
+ * supported by HDF5 for contiguous datasets, but not for chunked datasets. By setting the chunk sizes correctly, it
+ * is possible to greatly exceed the performance of the sieve buffer scheme.
+ *
+ *
+ *
+ * |
+ * \image html chunking3and4.png
+ * |
+ *
+ *
+ *
+ * Likewise, in higher dimensions, the last dimension specified is the fastest changing on disk. So if we have a four
+ * dimensional dataset A, then the first element on disk would be A[0][0][0][0], the second A[0][0][0][1], the third A[0][0][0][2], and so on.
+ *
+ * \section sec_hdf5_chunking_part Chunking and Partial I/O
+ * The issues outlined above regarding data storage order help to illustrate one of the major benefits of dataset chunking,
+ * its ability to improve the performance of partial I/O. Partial I/O is an I/O operation (read or write) which operates
+ * on only one part of the dataset. To maximize the performance of partial I/O, the data elements selected for I/O must be
+ * contiguous on disk. As we saw above, with a contiguous dataset, this means that the selection must always equal the extent
+ * in all but the slowest changing dimension, unless the selection in the slowest changing dimension is a single element. With
+ * a 2-d dataset in C, this means that the selection must be as wide as the entire dataset unless only a single row is selected.
+ * With a 3-d dataset, this means that the selection must be as wide and as deep as the entire dataset, unless only a single row
+ * is selected, in which case it must still be as deep as the entire dataset, unless only a single column is also selected.
+ *
+ * Chunking allows the user to modify the conditions for maximum performance by changing the regions in the dataset which are
+ * contiguous. For example, reading a 20x20 selection in a contiguous dataset with a width greater than 20 would require 20
+ * separate and non-contiguous read operations. If the same operation were performed on a dataset that was created with a
+ * chunk size of 20x20, the operation would require only a single read operation. In general, if your selections are always
+ * the same size (or multiples of the same size), and start at multiples of that size, then the chunk size should be set to the
+ * selection size, or an integer divisor of it. This recommendation is subject to the guidelines in the pitfalls section;
+ * specifically, it should not be too small or too large.
+ *
+ * Using this strategy, we can greatly improve the performance of the operation shown in figure 4. If we create the
+ * dataset with a chunk size of 10x1, each column of the dataset will be stored separately and contiguously. The read
+ * of a partial column can then be done is a single operation. This is illustrated in figure 5, and the code to implement
+ * a similar operation is shown in example 2. For simplicity, example 2 implements writing to this dataset instead of reading from it.
+ *
+ *
+ *
+ * |
+ * \image html chunking5.png
+ * |
+ *
+ *
+ *
+ *
+ * Example 2: Writing part of a column to a chunked dataset
+ * \code
+ * #include "hdf5.h"
+ * #define FILENAME "file.h5"
+ * #define DATASET "dataset"
+ *
+ * int main() {
+ *
+ * hid_t file_id, dset_id, fspace_id, mspace_id, dcpl_id;
+ * hsize_t chunk_dims[2] = {10, 1};
+ * hsize_t dset_dims[2] = {10, 10};
+ * hsize_t mem_dims[1] = {5};
+ * hsize_t start[2] = {3, 2};
+ * hsize_t count[2] = {5, 1};
+ * herr_t status;
+ * int buffer[5], i;
+ *
+ * // Create the file
+ * file_id = H5Fcreate(FILENAME, H5F_ACC_TRUNC, H5P_DEFAULT, H5P_DEFAULT);
+ *
+ * // Create a dataset creation property list to use chunking with a chunk size of 10x1
+ * dcpl_id = H5Pcreate(H5P_DATASET_CREATE);
+ *
+ * status = H5Pset_chunk(dcpl_id, 2, chunk_dims);
+ *
+ * // Create the dataspace and the chunked dataset
+ * fspace_id = H5Screate_simple(2, dset_dims, NULL);
+ *
+ * dset_id = H5Dcreate(file_id, DATASET, H5T_STD_I32BE, fspace_id, H5P_DEFAULT, dcpl_id, H5P_DEFAULT);
+ *
+ * // Select the elements from 3, 2 to 7, 2
+ * status = H5Sselect_hyperslab(fspace_id, H5S_SELECT_SET, start, NULL, count, NULL);
+ *
+ * // Create the memory dataspace
+ * mspace_id = H5Screate_simple(1, mem_dims, NULL);
+ *
+ * // Initialize dataset
+ * for (i = 0; i < 5; i++)
+ * buffer[i] = i+1;
+ *
+ * // Write to the dataset
+ * status = H5Dwrite(dset_id, H5T_NATIVE_INT, mspace_id, fspace_id, H5P_DEFAULT, buffer);
+ *
+ * // Close
+ * status = H5Dclose(dset_id);
+ * status = H5Sclose(fspace_id);
+ * status = H5Sclose(mspace_id);
+ * status = H5Pclose(dcpl_id);
+ * status = H5Fclose(file_id);
+ * }
+ * \endcode
+ *
+ * \section sec_hdf5_chunking_cache Chunk Caching
+ * Another major feature of the dataset chunking scheme is the chunk cache. As it sounds, this is a cache of the chunks in
+ * the dataset. This cache can greatly improve performance whenever the same chunks are read from or written to multiple
+ * times, by preventing the library from having to read from and write to disk multiple times. However, the current
+ * implementation of the chunk cache does not adjust its parameters automatically, and therefore the parameters must be
+ * adjusted manually to achieve optimal performance. In some rare cases it may be best to completely disable the chunk
+ * caching scheme. Each open dataset has its own chunk cache, which is separate from the caches for all other open datasets.
+ *
+ * When a selection is read from a chunked dataset, the chunks containing the selection are first read into the cache, and then
+ * the selected parts of those chunks are copied into the user's buffer. The cached chunks stay in the cache until they are evicted,
+ * which typically occurs because more space is needed in the cache for new chunks, but they can also be evicted if hash values
+ * collide (more on this later). Once the chunk is evicted it is written to disk if necessary and freed from memory.
+ *
+ * This process is illustrated in figures 6 and 7. In figure 6, the application requests a row of values, and the library responds
+ * by bringing the chunks containing that row into cache, and retrieving the values from cache. In figure 7, the application requests
+ * a different row that is covered by the same chunks, and the library retrieves the values directly from cache without touching the disk.
+ *
+ *
+ *
+ * |
+ * \image html chunking6.png
+ * |
+ *
+ *
+ *
+ *
+ *
+ * |
+ * \image html chunking7.png
+ * |
+ *
+ *
+ *
+ * In order to allow the chunks to be looked up quickly in cache, each chunk is assigned a unique hash value that is
+ * used to look up the chunk. The cache contains a simple array of pointers to chunks, which is called a hash table.
+ * A chunk's hash value is simply the index into the hash table of the pointer to that chunk. While the pointer at this
+ * location might instead point to a different chunk or to nothing at all, no other locations in the hash table can contain
+ * a pointer to the chunk in question. Therefore, the library only has to check this one location in the hash table to tell
+ * if a chunk is in cache or not. This also means that if two or more chunks share the same hash value, then only one of
+ * those chunks can be in the cache at the same time. When a chunk is brought into cache and another chunk with the same hash
+ * value is already in cache, the second chunk must be evicted first. Therefore it is very important to make sure that the size
+ * of the hash table, also called the nslots parameter in #H5Pset_cache and #H5Pset_chunk_cache, is large enough to minimize
+ * the number of hash value collisions.
+ *
+ * Prior to 1.10, the library determines the hash value for a chunk by assigning a unique index that is a linear index
+ * into a hypothetical array of chunks. That is, the upper-left chunk has an index of 0, the one to the right of that
+ * has an index of 1, and so on.
+ *
+ * For example, the algorithm prior to 1.10 simply incremented the index by one along the fastest growing dimension.
+ * The diagram below illustrates the indices for a 5 x 3 chunk prior to HDF5 1.10:
+ * \code
+ * 0 1 2
+ * 3 4 5
+ * 6 7 8
+ * 9 10 11
+ * 12 13 14
+ * \endcode
+ *
+ * As of HDF5 1.10, the library uses a more complicated way to determine the chunk index. Each dimension gets a fixed
+ * number of bits for the number of chunks in that dimension. When creating the dataset, the library first determines the
+ * number of bits needed to encode the number of chunks in each dimension individually by using the log2 function. It then
+ * partitions the chunk index into bitfields, one for each dimension, where the size of each bitfield is as computed above.
+ * The fastest changing dimension is the least significant bit. To compute the chunk index for an individual chunk, for each
+ * dimension, the coordinates of that chunk in an array of chunks is placed into the corresponding bitfield. The 5 x 3 chunk
+ * example above needs 5 bits for its indices (as shown below, the 3 bits in blue are for the row, and the 2 bits in green are for the column)
+ *
+ *
+ *
+ * |
+ * \image html chunking8.png "5 bits"
+ * |
+ *
+ *
+ *
+ * Therefore, the indices for the 5 x 3 chunks become like this:
+ * \code
+ * 0 1 2
+ * 4 5 6
+ * 8 9 10
+ * 12 13 14
+ * 16 17 18
+ * \endcode
+ *
+ * This index is then divided by the size of the hash table, nslots, and the remainder, or modulus, is the hash value.
+ * Because this scheme can result in regularly spaced indices being used frequently, it is important that nslots be a
+ * prime number to minimize the chance of collisions. In general, nslots should probably be set to a number approximately
+ * 100 times the number of chunks that can fit in nbytes bytes, unless memory is extremely limited. There is of course no
+ * advantage in setting nslots to a number larger than the total number of chunks in the dataset.
+ *
+ * The w0 parameter affects how the library decides which chunk to evict when it needs room in the cache. If w0 is set to 0,
+ * then the library will always evict the least recently used chunk in cache. If w0 is set to 1, the library will always evict
+ * the least recently used chunk which has been fully read or written, and if none have been fully read or written, it will
+ * evict the least recently used chunk. If w0 is between 0 and 1, the behavior will be a blend of the two. Therefore, if the
+ * application will access the same data more than once, w0 should be set closer to 0, and if the application does not, w0
+ * should be set closer to 1.
+ *
+ * It is important to remember that chunk caching will only give a benefit when reading or writing the same chunk more than
+ * once. If, for example, an application is reading an entire dataset, with only whole chunks selected for each operation,
+ * then chunk caching will not help performance, and it may be preferable to completely disable the chunk cache in order to
+ * save memory. It may also be advantageous to disable the chunk cache when writing small amounts to many different chunks,
+ * if memory is not large enough to hold all those chunks in cache at once.
+ *
+ * \section sec_hdf5_chunking_filt I/O Filters and Compression
+ *
+ * Dataset chunking also enables the use of I/O filters, including compression. The filters are applied to each chunk individually,
+ * and the entire chunk is processed at once. The filter must be applied every time the chunk is loaded into cache, and every time
+ * the chunk is flushed to disk. These facts all make choosing the proper settings for the chunk cache and chunk size even more
+ * critical for the performance of filtered datasets.
+ *
+ * Because the entire chunk must be filtered every time disk I/O occurs, it is no longer a viable option to disable the
+ * chunk cache when writing small amounts of data to many different chunks. To achieve acceptable performance, it is critical
+ * to minimize the chance that a chunk will be flushed from cache before it is completely read or written. This can be done by
+ * increasing the size of the chunk cache, adjusting the size of the chunks, or adjusting I/O patterns.
+ *
+ * \section sec_hdf5_chunking_limits Chunk Maximum Limits
+ *
+ * Chunks have some maximum limits. They are:
+ * \li The maximum number of elements in a chunk is 232-1 which is equal to 4,294,967,295.
+ * \li The maximum size for any chunk is 4GB.
+ * \li The size of a chunk cannot exceed the size of a fixed-size dataset. For example, a dataset consisting of a 5x4
+ * fixed-size array cannot be defined with 10x10 chunks.
+ *
+ * For more information, see the entry for #H5Pset_chunk in the HDF5 Reference Manual.
+ *
+ * \section sec_hdf5_chunking_pit Pitfalls
+ *
+ * Inappropriate chunk size and cache settings can dramatically reduce performance. There are a number of ways this can happen.
+ * Some of the more common issues include:
+ * \li Chunks are too small
There is a certain amount of overhead associated with finding chunks. When chunks are made
+ * smaller, there are more of them in the dataset. When performing I/O on a dataset, if there are many chunks in the selection,
+ * it will take extra time to look up each chunk. In addition, since the chunks are stored independently, more chunks results
+ * in more I/O operations, further compounding the issue. The extra metadata needed to locate the chunks also causes the file
+ * size to increase as chunks are made smaller. Making chunks larger results in fewer chunk lookups, smaller file size, and
+ * fewer I/O operations in most cases.
+ *
+ * \li Chunks are too large
It may be tempting to simply set the chunk size to be the same as the dataset size in order
+ * to enable compression on a contiguous dataset. However, this can have unintended consequences. Because the entire chunk must
+ * be read from disk and decompressed before performing any operations, this will impose a great performance penalty when operating
+ * on a small subset of the dataset if the cache is not large enough to hold the one-chunk dataset. In addition, if the dataset is
+ * large enough, since the entire chunk must be held in memory while compressing and decompressing, the operation could cause the
+ * operating system to page memory to disk, slowing down the entire system.
+ *
+ * \li Cache is not big enough
Similarly, if the chunk cache is not set to a large enough size for the chunk size and access pattern,
+ * poor performance will result. In general, the chunk cache should be large enough to fit all of the chunks that contain part of a
+ * hyperslab selection used to read or write. When the chunk cache is not large enough, all of the chunks in the selection will be
+ * read into cache, written to disk (if writing), and evicted. If the application then revisits the same chunks, they will have to be
+ * read and possibly written again, whereas if the cache were large enough they would only have to be read (and possibly written) once.
+ * However, if selections for I/O always coincide with chunk boundaries, this does not matter as much, as there is no wasted I/O and the
+ * application is unlikely to revisit the same chunks soon after.
+ *
If the total size of the chunks involved in a selection is too big to practically fit into memory, and neither the chunk nor
+ * the selection can be resized or reshaped, it may be better to disable the chunk cache. Whether this is better depends on the
+ * storage order of the selected elements. It will also make little difference if the dataset is filtered, as entire chunks must
+ * be brought into memory anyways in that case. When the chunk cache is disabled and there are no filters, all I/O is done directly
+ * to and from the disk. If the selection is mostly along the fastest changing dimension (i.e. rows), then the data will be more
+ * contiguous on disk, and direct I/O will be more efficient than reading entire chunks, and hence the cache should be disabled. If
+ * however the selection is mostly along the slowest changing dimension (columns), then the data will not be contiguous on disk,
+ * and direct I/O will involve a large number of small operations, and it will probably be more efficient to just operate on the entire
+ * chunk, therefore the cache should be set large enough to hold at least 1 chunk. To disable the chunk cache, either nbytes or nslots
+ * should be set to 0.
+ *
+ * \li Improper hash table size
Because only one chunk can be present in each slot of the hash table, it is possible for an
+ * improperly set hash table
+ * size (nslots) to severely impact performance. For example, if there are 100 columns of chunks in a dataset, and the
+ * hash table size is set to 100, then all the chunks in each row will have the same hash value. Attempting to access a row
+ * of elements will result in each chunk being brought into cache and then evicted to allow the next one to occupy its slot
+ * in the hash table, even if the chunk cache is large enough, in terms of nbytes, to hold all of them. Similar situations can
+ * arise when nslots is a factor or multiple of the number of rows of chunks, or equivalent situations in higher dimensions.
+ *
+ * Luckily, because each slot in the hash table only occupies the size of the pointer for the system, usually 4 or 8 bytes,
+ * there is little reason to keep nslots small. Again, a general rule is that nslots should be set to a prime number at least
+ * 100 times the number of chunks that can fit in nbytes, or simply set to the number of chunks in the dataset.
+ *
+ * \section sec_hdf5_chunking_ad_ref Additional Resources
+ *
+ * The slide set Chunking in HDF5 (PDF),
+ * a tutorial from HDF and HDF-EOS Workshop XIII (2009) provides additional HDF5 chunking use cases and examples.
+ *
+ * The page \ref sec_exapi_desc lists many code examples that are regularly tested with the HDF5 library. Several illustrate
+ * the use of chunking in HDF5, particularly \ref sec_exapi_dsets and \ref sec_exapi_filts.
+ *
+ * \ref hdf5_chunk_issues provides additional information regarding chunking that has not yet been incorporated into this document.
+ *
+ * \section sec_hdf5_chunking_direct Directions for Future Development
+ * As seen above, the HDF5 chunk cache currently requires careful control of the parameters in order to achieve optimal performance.
+ * In the future, we plan to improve the chunk cache to be more foolproof in many ways, and deliver acceptable performance in most
+ * cases even when no thought is given to the chunking parameters.
+ *
+ * One way to make the chunk cache more user-friendly is to automatically resize the chunk cache as needed for each operation.
+ * The cache should be able to detect when the cache should be skipped or when it needs to be enlarged based on the pattern of
+ * I/O operations. At a minimum, it should be able to detect when the cache would severely hurt performance for a single operation
+ * and disable the cache for that operation. This would of course be optional.
+ *
+ * Another way is to allow chaining of entries in the hash table. This would make the hash table size much less of an issue,
+ * as chunks could share the same hash value by making a linked list.
+ *
+ * Finally, it may even be desirable to set some reasonable default chunk size based on the dataset size and possibly some other
+ * information on the intended access pattern. This would probably be a high-level routine.
+ *
+ * Other features planned for chunking include new index methods (besides b-trees), disabling filters for chunks that are partially over
+ * the edge of a dataset, only storing the used portions of these edge chunks, and allowing multiple reader processes to read the same
+ * dataset as a single writer process writes to it.
+ *
+ */
diff --git a/doxygen/img/Chunk_f1.gif b/doxygen/img/Chunk_f1.gif
new file mode 100644
index 0000000000000000000000000000000000000000..d73201a1b2e78115f3bc61964d3f6ac97325ad48
GIT binary patch
literal 3664
zcmV-W4zKY?Nk%v~VZi}m0e}Di00030|Nkri0002N0bv0E0{(=LsmtvTqnxzbi?iOm
z`wxawK$hl-rs~SJ?hD8AOxN~}=lag~{_ipXhs2`sh)gP%%;rt#j7q1}s`ZK`LbKei
z_X`e-SKG1qj83c9tTOuzkIUy;0R4{7>-X-x{|^{g)hAeJc!>C9m&n-Y_!vM4IZ0W`
zCTW?exp{@j`3V|YCrVmsy5o7Oy1JUm`U=bG8cSR2IBT1$3y8bx`zzNAJWOmRT#THI
zb*$X%%*6aGJq;91eT{8noz30dw&)!$Ub8)Jj$MAP{+zzeNn39xFEzori;?
zo~fv~=HM8$saU3Dp^i0M)@ib^Y0oYK>o#uNxlc#^(27-)S0#8$Rn^EAt>D0f(<6;~#qR`rI
z4Ib!Vk9EP**^WyIi6W07>UWnl7WEgSA~k+6sWR~`b
ziQ$V8Fsa6LBf#+FCONY80GM)GInywT#R(aab6%OGgs>?f%Q_1nN2V2pPAS)*bnba#
zp?W0>Xnh?r%IKkFwuvJHZmQG3pN?LpVUnAbsOX}m8n~8YnlcClpROK}BXw?IiY6E9
z^+hX0tz0@ORHM>F;;CWkHeZQ71xg>z=+42!%`|NU3OlJl%Cm|
z8lk9`c_A$~xu#+1hTOvU<+sKuvn){U#*1CL;=*w*U(gOnZ4>r7i|u{h#+z!urnS0m
zk>aV;?}p`?^YDrh%f>6R2{YB@#YZYgFGw2m*(}2o>#*j&ZTvTKFcllj8=f8ev@OfP
zA}lk=4Et$bq|2Brv40^`rR>LJDf+TdDKk>EQbIfUugI6I+6m7pwPwTTo}c
z-O>Xu4PmGNr#WlG=Ayi-x;3kMEY};;<1S9~g8kUfT^%j;6nN{ZinVM1{cX%~!}K-a
zb>Doi;Cb`CmDnP?4LO2)AO4c`+=j8?a}M>
z`B!EmGW}ZHCy_mESkp~D?wh}S`t@%kp8W2AOwaryh{_)PgS69Yeffgf4SM(P8v(!X
z;R|*X0gM4}bHD;3aDfVJpiL6^Kn6z8fgWQC1uZzi3|4R>7v!J@F}Ojc*at5f;ox)1
zP{K&1kc2EmVG9AWLKn)=QL%#6|0L6*-US49Vagx<)Ca`tG$w%lLqiQEBfwf=hKOmq
z*Ok(zzaailJJlIP{t};tGbF~(JhoVpp{|%SwH+v9Q&i2C%2+?_Ekti%BwH3AhbN!y
z>P$NH%gPqzvMA~ijZbtyOro&GMjX*$ZOlul)(5^q8j*)`BHA9)2uZFvl8LW-qo{Ot
zLnrBslD@m+hqm@eDMr#mP_tqYt9D1}q$H5dg9WY>$;w2`&r0LFNGP*&#!>RoA#MBO
zDUFdz1-UPl-1FoucZp10g3AfSq+Brn7!Yc3a))A6S~4pav}+#IXm!lYGn-ima7L4j
zzl@eKF(gV!7H^yBtW!5dQA00ksf38!mpBD6B?&0AV`SVXI}-#*MWT|Baiqi=cX!V~
zv?h<*q!dH`{n^m{F;k$x#MvP~anaIAA)&lrq$4A`tdBx;q*Qt*g*;hGH>GrQ0WGH(
z-ULdJ_OMj#lxZg@YEB8g(xtu})2~>^(f1Lsmo`0BErpQFnv%4qmDJtv5Q@{6o^z>1
zRjN{3+Qjw&Fq2D|rX8zE%4||~tX9RGN#|)vRdN-4XT|7L1-8zr{!^|nYpX1tsmiEo
z;eSCz2_wh2kg{TotaGhrB6$jiK|ZgbjT@$36Z=*w`jN17rEJ*z_1G^c)>-si>s{=s
zQn2<&q=L0)n?I}yxW@47p?XK3u>RqpTbGgpxt*x|ciQjp*
zvT%y4baA^;=XPPUzd5XYt99QjaM!lCOe!t|%ippjmzqxX1r7B%Bj7&QPwRazSsvP8
z@xl|Y4LKwa<$xN0=lP&mU6FPayPzDJELwIEnYT1KW
zmSC4VC_*o9Im}?j$R@MQ(miL5fd~$h+>4H;m^rOUhz$4WE-K
z-B>x!>)-iwyk01x6Vb4u@Qm+FC?yU0i*T+5j~!gUOB)BkN%7%W`KP@i<4dDvItiO5
zZN*B{;>X2>Fi#ZyY4+I}(1NtEndmHQAs;%=f+cNJJRL+o+eXy1u0uwrMQqp*Z}7MpnG@e8y)AC={;t{@m?>ZQ
zzK2~fX9vl{AQkv7g-xiO>-^Z}>^XZF4s=LAd@#^HX3f!9aYJ7m5<68kj3>--R|*|9
zM=gxUjau%a{yfqqX}jCs4)=Tnd9iV?yWQ`OcVFlE=KagN-~SGHlHeL>Xqvj=V{R0r
zD}C`)fx4g&5BZND&+(ihb!8N8(6R4W^Tu0y<~
zJ@Ru8ee9#w_SEmb*ST*}(qew_LI3;ZXMb(PAHVUzkCo*9zjXQXmoxd@UwqVi
zfBz>t{^!>Zfc_?X{6~NTI1a^!fTWjo^H6_@MStp-fRz$}8smQl=zHFr
z=7Ajuf-$6INXCI87-c3{f*bgOD7b)q_8{
zfIujOwlRc72s%cHghZHxODHQ$=!Dbqgi$y(IVgn;NQG7S4m^m3p|^!tsD)iECLqtUhToTAXedAMWrySdNT#6;He^d*5nFka6Im8RC$vFq
zML2!fDqAusl#wiqBNZLQb#!EF#K$p~wTm}}|h+<+Gg=DCX
zoKlPJm<{X*j|cdU>H&CFD34MScn5(-B9Je#kT;f4sFIOC0!;`hjGzdNcMy^aMUsEv
zfR{#$HEEMtmW%k5koaYhGHH^E(1#ndN)ri9DE_IFBiWOe5(b{g6+?-Pff$i~g_NkI
zl+03n6I6&T#5dg?}wIcS(XmDcrV#e
zN%@IC$%S&MmUO9fb}3YNNtb$QmR%&4b8wN&R*?Gg8;6N+Zs`${!xw-#nQ19iVL6$O
zDT7%EmT&nTOV^l|xs)ZBOTr5Z}3Sb9@hYM4m*cVSwQ
zXF8>6>UU^5k4*ZeNE)YbI;RXurggfTc>04@s;8EAr+tc6fZBt6DyXkOsD(sQkyMj|z5@3aNiesib&4soocr1w_vHO
idX%nOm#qq`LN%*@DXX=rpMj^VyUMG*>Z|-v0027{BspyW
literal 0
HcmV?d00001
diff --git a/doxygen/img/Chunk_f2.gif b/doxygen/img/Chunk_f2.gif
new file mode 100644
index 0000000000000000000000000000000000000000..68f94337d86742bc6c3891dcc6e6e6e4187abcf3
GIT binary patch
literal 3986
zcmV;D4{h*ANk%v~VPOFw0)PMj00030|Nkri0001C0U-ha0{(=LsmtvTqnxzbi?iOm
z`xAuXNS5Y_rs~SJ?hD8AOxN~}=SI%={tpZahs0uPXG|)W%%=0H13;(Ls`ZL>A+y}B
z_Y1CV!{oC0Tpgp=?6zAheaq+cy1I_f>$CU0{|`vi7f4uW7&CaNxX2ig*a#V^_b6Fu
z$wqmpx#^Y3dG(?2Y34af73YI$`c=C6q{;*9$|pMu3RG*Gma99(E7r@^8f^4SV@$j>
zoIHdqNf3o$CD^Yc0++U49eJqK*#Nu1x+e%`T4^UoUiy4?};n
zFWB$jihw5w3Phpr;J|wdqcr?sutda%A`((GF>ypj{)!ealGqpmWW$agD>5wkOr%1Q
zDi4yJ6)FL)m>>kcbk*`qPD?je?JVV`oJsH~pqaJGXN*Xq4*+Rnfpm2FnC7
z_-{zUjMEw(Bo(7xm~40T;js9lqs^Q-3s!tLb7#3*K8qF|$23<_l38m$S{aIKjUFU!
zPQ6w2=-s+sd%g?HG33dl)6(WkDt42|#o_M0{X4Vn;I`|c9!*{CYU9Vbhn(G3xjE+D
zD}V-my|b$9$Nixm^VxLo@8HXNr;i}1`0@4r2~UXIUQLh9iBX+Iq?>QB*`=8}!5Jr=ZSJAwn-EQ+r-o#18R#~6qNK&3y`YJx
zR*lZU=%Xn~3hAMgBBbb5mR>3ur$+thX`-QOvFRFpwo#}yk>YV`SbWNfYN=MEdi
z-U(`&w;;YI@{wK%CeYP;XL`%RJTb_y>sMQw=cy!9R_U8S||Ym0n=y*uz+r4hU>
zzySjmooBc~H><)6`}T0d2L3y(vJ$5!aKHWPTNbxaZoE!zCVLzm$OA_^GRXrKwu68%
z%Uqhi=Mp-u%ZsYKj-FD&pV!7klI}%9Wl#dujIDbBG3I)*`2WMV9RK?t@ns`^Br&9
zdk_A1qJh7Nc;cJj&5z-Zr~Wx>x|Uy#dFGm973{H-TcI-Kp?xk}+}D!MFR#y{{)*|p
zvR-=UtHTZlkD-T7I}WYh?)edtN4>l2%L?5z>%q%@y6=lSeZ1<)W4e6O%_sak;Kidp
z{j#_p{X6rt@;-atqI(K{_2LumQTeauZjSIfvcH`BV4x4M`oqJIqVwzHUnTPU^DhYb
zRRw^E_?`eeLBImq?tm2{p5L%%K>yuOdDkOd_SV)r=vDA@7o6YrDEL4NMo@nf)L;lX
zI6~_E?|cC4pxwfEA^LHIfx>X$_`pR&J|s{YImCht>t(<9?XVd=yu}dnQAE%k5s4BM
z;u3SE!V*5wdM6Cx{`{uMy#+S$ido#?RHW_@nL0k1=ea
z88Jx5QO+_Mx;(`#d&$RNdXSf6G-RVn*+5o06Kkpz(kYKo&G=n2MU3R)Hb2$POn!4J
z*=*l9o5IOxo|ByEtl>3_i4A5t5}rAdXC0?04tl=gn9Urfy_ESv^C2^r01cx+`{_$9
zhSQXQ^k*>s^ob;U3NxV_gr++S>Ck^zREN^csMN}tzjbD1q3RUr(Lg#Sla{QV995}%
zNP5x7HAqz`)et0L#h!=Ow1175&xxL+2#79+pCN)6l-84^doda=}&fR8>h~J!3Gnx~2ia
zEU7gUOe?chS709WjUEkHLEAXkdlq($P7NSb-`Q9qy7Z-D_2y*FT3K-l#a}RqEE5kA
zy+wJ}P;EijN=93*IpyYH17gr(T6SSWhl?Kd^yRQ~OzcGS1)l^7x34p+ZE=ko-OMtS
ztSwsCaRa)-fdXQ&DS59!X9&3qej;zLw$k<;2!sYzJgS;S2Ubf_)1a36pkd6t6WJapGLt$
zo-CF_DasqCEyPC@^EPt)U4jO=yIsbsn$yeWDa)6AW}XR+y?Dp|`ti>DeVd+P$^K`2
z7H-Y{U9gFF_~lh%beoB`rk1xHVL3mz(b{}8rOjz+C+pzRkmhikReI@=da};VBJ&VQ
z4URxF*%YW=GMe2CzgBNa)jKAorZ2f@Aj3J*W5s2x|7=TI>zc<#c5tqfTZ8{UfccDAuu?R6X5;S6{5wk57?
zievoR7Do6}3q0_Bni|m|k7&Vboyv+=TH$mq`B2e4?_GOZ(u76%NY%Ss{$ID8-he*&
zu;0D$Zp(Yr59ch)HRZEopIZ@SYdP;^cQ)!xkS+Z<4hbN4I*Uh|*7eePo?+1Gi!-MVk=?lV^_zr#K$wF92SgWrqU3t#HI
z1|IQw35!ox%J{YLthwlnyrYtb8<@SzV8ONg;>}kWjy;S(dDgqg^>|&;qtNswe^$j_
z&oQ9-ne{L49qhG>`--8B?@|Wrf@tO+{Df@xXdk!dn?-x8A75O57xMC%)_gWYH~P;f
zi{PvO?@>KH0`7u4dZBJM3s~hE+&}vR1WuU-;N$%30vhkR{{Affg{Syii+|K{oP6@B
zpY78>{3kd!CV4_$)F(ZzL4PR*SczbN-UA}}r#sYQCkwcD=0Qb$XH4?qfZzgt#3OzC
z)DaRmf&URX=Ta#5=2kl=7PX@sbV7jn6<`uTD77bcIRS!pmxAI#csRCoPbhiVw|``JdE8WeSD1017dhA$clG9l
zr1E{d3$4rp|Xd0*fV}uhDOK}qQect
zz%DfyV}uxoWVkkc$cJRs3iE=9F$Rep0)|V0hmd!PtdjnTH@Asp^@md-h|ks`rTB^R
zCyJypidQCk^s-W`IBA{MHjD_1UKVApcyqOAQrH!Wnz%BjICG}RPCo^V8Ma}wIE>a;
zjA?gaFT;qqxQsFgVVLra$QX@NNLk!djnH$AL+FdXvyEBQjWGy_)R>HmIF7|)blP~1
zBdCjD^M=G|j^9`g5wyECzXorh%=wqj{s?np7?|Ih!O*NhwP}2Y14wE
zrjY9hjsz(v4|!@4nTF0tkwb-%F}F1NXpnx?k;FHWq}7Wd8Ga=xF_ZXN8rhJ^;)n;w
zlF@dO{x~>hn0II>bQAfKbR~rX$Az@ji8cuvP5wBDM#&RHnTm3QCp`x}VB8pcFcv7W$wVnxPVUp&Sa58_J=r
zXrK)`q9JOcAR3kLIie9dd=q-2E9#;xilPPDq8<98ADTY#h*SvaoBBDOGRb*N!5U`4
zB0j2`?m440rCYH#QpSL!lBJ#%S`MiP9lAFbq<2)SF+Rfobre2z&XE>zd
zrB@nM9;37%VI!mKd80w9p@~ZVp*8BJm1?LS`lT4EsY|M&h+3yN8mg4asU&)-rAnid
zN~Nais+L-*r~0a`N~)X6skADquzIVa3aYXCr!m^AsG6(5+Nr!crMqgY$r`Mq%B;t_
ztV&9(vudoWO0C0MtHS!M){3msnyt?2t;(vc;o7Xfiml=bu4&4x(WuRp&imnh^h~5yd-=wc{n5xj~sSC=WG=eYJs)p5aqV({A^@^}nNT4Kf6LLyr_?QtA
z>tz*dBm25~i8m7(3y;X*A{IMOKM^DJL`#G@DY9T1w5R8jM$3mfsc)b)lN>R%9oSPpnUGk^
zU+Jf{irTgM0hCpFlwwOvH`_B!sfQG(QD%FPR9UrP`menew>dbsK3KPQd$&iMw|2I-
sdfT^4__ur;xT!X{gj=|?cDRU}xKy~fdEvH?3%QXixsyw|22cP1JC=t4l>h($
literal 0
HcmV?d00001
diff --git a/doxygen/img/Chunk_f3.gif b/doxygen/img/Chunk_f3.gif
new file mode 100644
index 0000000000000000000000000000000000000000..e6e8457869c5f4b9011fb08374441b78ac0c9633
GIT binary patch
literal 6815
zcmV;Q8eru|Nk%v~VSobQ0e}Di00030|Nkri0001h0^k7v0{)DTsmtvTqnxzbi?iOm
z`wxcVNS5Y_rs~SJ?hD8AOxN~}=Q6-o!1iE6XwI2Q5urR2vl-iWS@tk
zwDHwar(>P@j83bOUCP!9CtAEy@x0lgvoCKU-q*G#sG!F+xQ3XhxHxhZ_VrT*Sys37
zm@t_cLbNv#%)I=HwJJ*up}Pk#AsIrelJ)tP%hM@?$ii0s8ArL2KFQ9?AW8CAt|vk`Ye^7qV}TR%9-hj
zE12XsBCYnasZ}the5x9XHm%yVCzA4GD+jLJxpeEIVC4ilwCK^KOPfB8I<@N6tXsQ&4Li2%*|clh
zzKuJ#?%WxXt-%Zfc<|rE)d+uayhU;r%g^{WUL3mgG|g31w;mnM^~;e#<-mIx(e}34
zor^DTW_%*`iM!ii-yVHC`SYi1hYMc>_#65D_4g0)9|!?{Fd%{Sg~wKb9PIbrZwht>
zp->Oqys6$9Dag&X=9{n^NWMxlAgY`mV6;LP>9rt*&V9uMIy7C%X>`{Ow8OrCTwNZQ2VUeB$nuZ^#F^`?0q!
zf3&hpB{vpJSpp%Yr>Fp*<(bG3&n#`s?nbQef`ft#EVc!Q8IQyyAym$gRzqa*wTlzq!~U#w#2ll0UCV&+%Q|t1iCM(lTuY-RjFuzSJ_`KmLcb*RHVu
zZjByqbDRmtwl{6XWqPwa-{Go*H}*-;e3KFhoIv*r%%}~4?OWi!^u@mkN^pC55Sks_
z6vF*^uzb-I68e@GKM~3hf1R?R|3Y;QuJMn0_JAP#REWFV1#yT13`!y1#KCe&?;OHY
z7E65iy%|a|h8F<~{t2(OL>9VEfmGbv6{pC#W5sDhEMi>QlGr=f?WuuX1mpU?$h6&w
zX^vd5UlYMczZ9mihzAoQp8n^)!x0WSMPgtJDL1wcx>1I3+#ecq_bo$mQGOKE<0328
zNH)feSvB$(?;iQXJAe{?nY?6QN*TR`*{O~?yx-442n_=n?1ljRV=avs%HedfkOm1J
z9;vrSQ;HCklk|%-A_$J^F{_f66ehKZxyAYX=$H-DTn7=!#x^?hX3zv75wXKc?^x3!
zC4^-Hv&lzV`jKGk1d6>zxy)z9Qhh&z=fb+VO)*w*Fx1@EIL`?~Z_+TIYs2U28Wg0y
z%~P2Ivu7**2|7W8>W!Urq*L>>sj3~4Goat3Xh-{)&KepCYUEsKF=^q@m_Rh5`V5Rh
zJH}A%9a5VsU6n-php~hv%A^84T1q!s)8yUsnjXbr`FQ%ea^5kR%(P`YLF&|xYGhvj
zBu`0U8clEg6re#Z*-(M%Qh8#vr{&BlO}m<=uW}V*XO-#8kb2M-!Sgw26>3XEYSp6}
zE`0HfX%tTi)uKifuQO09Q<>DvJen|Dd@bu-aavc&(B`gT?cF@v6(GOi4O3d5;C8JX?Vn+*RH)a?Ql$7oQdqplT(_z=
zx6zd@V!oS8b@t4*1dK0gho-@a?iR7PG}E+D2H(??G`VR+?{h&m%6_)jt&*KBU@lwV
z)0&E`_g%1l4-?=D&$qYeEZ}yb`?UmL*uoUXu!H5h;bh*{z8pS^Sp_^|;P#h7RQ)YG
zY8+e!SIWSl#c$YROyVB%^Tft1@N$9d-oxfN$Ss~rhL@bzCqWn`(B+lQV*F&zWEqRh
z8}CNwOXXZ<7RH|J?qJXSop;)J9XM|CBt5Li{%{1yTh2_KmDptypVze9ZEk4qJpNVE
z2$wb5BeHp=Y-F>jG<0M9^CB^o6`LBmu>7?dZWoQ`ZlyQLC_W%)glwqlD!I~9CDcQ@
z(`H2bN^{v|v>d4j=)|dnl3*0vIIS7$H;?s)BHHw7D^rrO<@(0wG2pIQJYX4LRGfXD
z-K#ZyqGLu%n6`E&m0fjT3RJo7@Iux4YB*Zmx_w
z-0@zwi#tc}eE$S`8
zO@7Ztg=;S>dr9Q(fTTAXb*WscSgg0aWa}mf-
z{-n)o`xQUG`QohE^s&83$v<`WqYr&Lmap{^AQN6Zk##&it>az(ozPRA5Vh}nk4uva
zzm;mOfAN1;5BcB}_s0mz*EBA-e=4PS>DN%Wph1koa0FOtZMSB5rF+BIYkD$q&7ewr
z2YBOGeghOl2zW;P=X0Rf2zMiamF0IgCrK8S7oxF(LpCP1LxP{=B@uWzGblv{=z;E+
zF9@WAtVc*cxJti=W`X5S2MAogA$_Jtd^}h$*LONf=oh~Vj%c|6BvTAhkr|GgG}OuuQzgGD27@_g?uqQc6f(fMTWofJmP~y4A_T?
z@`q{&PV3i!WvGPY*M^PwhEv6aT_|K0$cKdpY0iOw}z|;8IvfAm3Vh3Xo}#aiF(0_xrm9NcrYhJD#A#JzGyMKXkC=H
zey(_Qh}ewINQ%sOcC@IB>k)It2qwVTc%6uir?`c|h)~v8jNP~>$Y@+-#CxqMi{~hf
z(YTJ>M|w^rjM!5^WGpLS#&Rn9zE%kt_OUN;gnBF
zBS4vZ0XcUt8Ix2QM*Vn{$@q>?$dE{xlnEJ$N?CqWxoTXgi(bi=ZyAqq36WB{dvHmY
zZdsNrnU_YnmpK`hK{=MQNJMH$N_NRAR%w`t8JLqdn0GmjirJOl*q17~l6+a2emR*$
zxtPB2Uv;U6o(Y+ahnJyAngxiMmxq-;0-B{M7&vK`uL+x$iI94EnUtB5hbKgYSxu~2
zXrq~hx@nWHshL~Jc&W*opXr;o36`0;lg-JFe+igh`J653QJwiszS&z;)ScOxj@wz3m|2@zftu0@kL)Ry{_bg-jR~JuX`JL)lgT-Q4rQOXiJ$Hmo@X|m
z`$>-)f@(uamgzYaS$Uqeshrb^pc_V4pO=jPbuJRRpC2b31?rqk_G}l*X6HdSxq=D`
z7jOla5Ccb|2xmMg%A)$_ZzbBIFq&{PDx)>3qBk0&I7*{BYNI*oqdWSe@K+%?7uj;Ii;L*ogO-qpQUmFNQV;Gp;*eFlBt)fRH51lrcQR6
zKWU^#>Qf5Zq|u3_x459^*HL78rDy6sSfi&`!>9duf6Hi!dn%}W3Mhm6N@8l6b}F8D
z>M*kDrgF)g%lV)#kx>pAo3E(;GPGu>*5-c2iGPX;pp2?C!}X*Kw}^cRsJ|E}o4Ti#
z+NiU+oS-UL0ZN|#p`U8%s*c*A+Es^fimS9!rnw3Tg<7Y`Qbe?xotk2QPRfHN`I{J*
zreIi}YT9%&rIVuC3*U*JyV{2!Wv)Di@&P-x2{W@+zKsq>%7Fm
zQY*h{>bGnvTK$Wk$=kj=`>srDW)TcT27J5~oWO`0V%s~tS%;N-7ru$*c1xSVf&0HG
z+yof;!P`2Be1=&jyoqEhx^H`j8cTiEYiI@-J`h}$>szvh3&QW?DnM7isyVDQ?2|Yw
zskqy?Us-0eE4==j!c*MD{F}W8yMQx;!Hzq?_ld2`I}euPwPowX8*If(XR^(~xA6Cd`!;`zf2Nfk5XrLCCxBwi)6pX+EOTMRTM}O54qMN@jQIZpad_J6wKC`u|
zcPMe?{=-I$#&ui5tJA|C*+FMWxF^iP9tyoUwy*kAxE|cXefz}J%f74o$F59Vk?T8>
z$$5#qv2x+a6&c7)NWjzv%G}bb7M06hMZaYnb}NRyQY?GhHK<4XhhO^2f(*OqcfVbP
z$e$d|(oDr!+`yvgcstg*SbWL(m&%pK!c*i+Dm%`~yg~xJyN@@KSQotOT*XPNv~~-~
zU!0zQV!j&e$JwmH+zh_k9E1A;b3)w5lyq?Pe8uNWfN8^2!|aE}Y|Kx~bPS9g3(e4O
zOwKP`wXk~4Iy}wTE6^o9(vl*}TDWQ
z)H2M~=IPBtyM0)l)j*8aSUR9ljL^q2hhVnWzZ)H=mdp)Z&v|XqUVYR_-PM1sig9hd
zGi&BTlkM1&UD=fV*q5!@na$aj-PxG!*^w=zI`T;a
zXu%>V*q(dXcv#m_n$!CW*mzCSd>zz6{ne)|CtLbKy^BBYan7qf$Ds_|XG+yb+{S}D
z+-PaF9&OvVJB?eaq&}
z(aUY!;C;Dj?cSj&-e@Y(@jb-EMzz$f*M2SAdrjN>{onYFGxcrX4$PSaE~@j5%LI<#
zcTLu)?9b}G%hKK7{$1e#{-G(t;0?~T@yX$At<@d<;U*K}t31?T4c+NY;TJyLD(=wy
zIO4|$-!V>=3Qpr0Y2&yZ)?zK;`rP9_zT*0w;Wu8ZB~Ik3I^##~okq^i7S7^N{?S1$
zuu&|e&Vm{kpYUx}l>7)M8m~QF|R_a@Q=q8@xWuE7H9_u?1>KQL(c0<&gIFTpS#ZN6TR%RZtTat>C~?2*Z%3SUfs`b
zkG{_B?d9#?u941u-;D0*=WgweF72V7?R+lo#9q34{_lf~@NORM6>nh*kMX9Y@dJ(R?LP4z
zkM8LX@(r)t8{g~~kMiTL@+%*&if>`BX9E*{_pZze(yWabsg{Xj0E%8
z9P%PB^EhAh3}5UwpRGZUvFiZWo5O3%Xf74vJd>RU;D%VxS~JyWdBVk3Dbw`BeRJ-21*y`jPMaVgLR9r?2zb&+^!B{-KZl>3>PvKlrzA
z`Q4xV<4^v(|NH^Z{_|e^`9CA>&;KM&0P!((x&2`SkydNw)jMVWH4hvikvta@T`|^0
z)R&B9kgoCS&iUT|fkEMrSTr6#JETmxbUvX`>6BWvQkvJ0mfQ7y!C~<;J7$!RXrp?~
zWfR}SxZGVgZ{~OXp5OQXQS!zF0SgTe5fiNl!x$AGAtNOx@hBa6E^{(9?XEU!K0!Md
zEh$Y;QBze@N>^Q9VPo$?J!NfgadX=yX0{{aRRf{%_Cf)NfP
zbQdMzgN6oafN0z)o
zu?@lZ{3Z=6<*V7vh$>GOJreS0)2BU33XPhzzSZUkbIu$%HpAIsYae>un)lAqy@5~O
zePMWTr0RgyS!1UD9%)kW|91Oxi4xr&d
z2`99W!w*4>5WPJzXz-*8F+&iD4Mz--zYbNJK*I)UtkK3Bam-Q29eM20#~*FW&PW0@P}~wy^hD+&&2Uk0
z8r{L9ElYF|PC5xK)X@S#{uLFbP2(K3jfZ5c^V36B)uYucSbg!kK3R>mgF6{2qoES%
z^wrir@T4`=1F2Jqn3;*9I1AxmTvP>$Ksfnxk3o&Rf#I&E*;2wu^pSXrtd|I|e>ABph(F
z@lN;I$ic2$=@R61d|JILGO=B?x0ZYBz?o&+^sr4&9dv3tKHJ^86?Yx-yCuI=bh>YE
zA>+PJ=R72sn|8bIcvaW==WeA&yx7Dy$6NKoHTO1n!JXfeW^r>LJyq#v4*F!@!4F^j
N@yRdWd`b!c06RxY2j&0(
literal 0
HcmV?d00001
diff --git a/doxygen/img/Chunk_f4.gif b/doxygen/img/Chunk_f4.gif
new file mode 100644
index 0000000000000000000000000000000000000000..76f099459faf3a0c9e579256f78235b8278e8be0
GIT binary patch
literal 5772
zcmV;77IW!GNk%v~VSobQ0e}Di00030|Nkri0001h0^k7v0{)DTsmtvTqnxzbi?iOm
z`wxcVNS5Y_rs~SJ?hD8AOxN~}=Q6-o!1iE6XwI2Q5urR2vl-iWS@tk
zR7k9Tb>y=7jP_Et);7_iMrzp%`mz(Z*D?Ajyz3WON0$ePc8Hj$cycxN^&@sf#pFbF
zwa$huhzHhSB)_zN8T%k~*tT#Br<
zx)(}$8P@pRbWEB&3$4stnktC9ZQbD8?FY#PRxXDwyw2W^*vRQ=%n6!oy4);%pUWMt
zUzP2fAZwsb#q*~vAF^ZHptPz)DBY%By24Ro=q{ti{*5rMG5k2toV-vA?>I#G%o(O9
z)JUQuhEicbkPXL43}_FaCs{UQ@={ohqtKy5K{zrfvz4}bM^Ew-QVjGp?2h3EOYDHy^A+*S-pJw`uz(yu;9Uj3mZO+
zII&_KaThy&3^_78osuhCzKr=yZH-6OPfB8I<@N6tXsQ&4Li2%*|clh
zzKuJ#?%lk5_ZA8>&7sXLh_5iN0(lDM%ZEGX7+mvn&CD51H&flZ@a-D2hGa|`NA?ol
zZ;U5#enfip*UPNOo+iBc;qq;~H`?BPe#UnG?|Ue3AASDev0n}Y{#T%Q{l#}65W>|5
zT7-)=I3INsM)BZYRaf)z?vmxh0+!=H#g+$3R&keRpxi1sMRViP2(NF!n~b|f8%
z2EJIKjy3)`*o`j;`G$|?8CgRyM2gg*i%#mtp<7Fy!();&_E;s5TJkmJaZP49gGvfc
zwPl$fCbuAz7@gT=O5d3YCyp$%*_@hh((u)ew|oiZoTD+xrWpqA03*bi2#xdI!YZ6!KmqE5k;i;Qvs!Rbcwl{(o@27#p72dJ`VCzx{TgP>L-
z<*C7uJThG2Ua!Ztf9o}
ziK`K+L^`UrYcMMihWIcA;03)ZQjZm-VKa?x$%=sLtN8XwD_xDc>r<|9ASvq`I~lAk
zEBY#%UTQ=&94%Y#x}fBpnp#}0#c@sQt~0t$3oyA?!MiXp6Z^}rl$?_CXuyMzX_Y8q
z-CMBBq9pXPXwoK3-a95Rck--n1&t;-)b4^aGh!+YptCj07!=4G)+%L?WUPWW&O*!qa%ME!(hx?8r@K;*z
zWr$T0Jo`V1dn`@zY2U~FzQUJB_=&?)kKoY&BE!NCn5IebN@`T1#T+VUj?KkS4@+a81*5d7H{A__#aUnZHub>^iV#l;)Enlm
zg}cl_r++LQp$i9CysdRlf*3TQhXPc?Vns|(F$~G4Ci1fu8YYA4-1{u8h0Ibu~1dPn487vpBb8y@S6Lt)u`(x)FZV$p?aRAIW-2*)_K
zqm9ISW0-_k7%R$gk2~w(p1`QX9gazmfDGgl4GBe~ov|}|=1C44W#O3C(EM5t%vS%rd1J%~+<*l$IPOH;YNkW1=#g
zrb;Geu8GZNYV$vGhpCkmPH}?rpiMi;WS>#_mD@oAk
z9dx10Oep>Sbm&mU8I+-pjHv3m2~Ki`GN8c}CPwuc&pugnPV}TGB}eK^ld6=DC_P~@
zTe>@#t~4()U1df&+R=Y{^rkrdXg?zf(RogErfaNe*@%kNqJAr==PYSb;b>H+Zgi(S
z&8biCD8`>^RC)syYA>J4#-)OFtW1pRP)T}L!Jw0@STyTbcemAo8dIzMq-#&BddVia
zbzf;^s|ek?SHB8&uI1FLR|7lPIf|94cD*Y|b!yeCPBpT?>1tE^npocfHnW`7>|#HJ
zSV)Gpv%~l-WKWA)yh>KCgq>_^k3=??9&5DbIx1X@CzLnXVm0b18)$Z{7G-
z1V?DI{#5O1g;rPBzP7QHU7=ww17HbXkisG^v9C-!+sjJT#5E#uT!-Yd3P1M37@b;Ut4Fpz2FV;c{-wXDr>lbu{+#uj#;4|ZIHk6f7~V|mM5
z5^^Z3EXy5p`NZ0j@r^gUWGKTpt7%qpclZS6ZN_=bi1IR>@BA=3V|dQ^LvEKw%ABVR@ivT^^MOQ1q}($AE1-25CpOjDXbWmfc>1%2olTX@uGCiRpJ
zsp;mWdeab2B&%P&Xgz;g%z6nitgR{QLU*{-qV6@XeLd)1zd2>A&Rel>jZar2d)e^H
zwJooF7mhI-#Lk|znypRfC<{B(zYaEzQT^ZLO*_HW9=Exlt!!@|TiokDH$1<6ZFFZF
z+lJQlwzZAsVN*HW?{;@_E$#1s<2%p9=69n74j3)po8I*vxT(3_ZHP;ps|C-pxDoDX
zfMa~)+R->&+dVsw3tZv(ws*-L4(y4W{NXFFY{kiZaE_ztbg9dp$7Md|{^QlR=C!@~
zmks`CocnU;M5j5lW9;cfCtcnPS2~=F4)sPNtW+c)w8%9m^uts=>nE)h%^I%QfL@(P
zV`pin1HE*#g1Y4qXS-Hcrmw9RdhF8XdXb_IbvEnU?yoF6t(z@&i}#!Fc#(VCQJ!+Q
ztDNEBda={f%~QHxobWU)H|1aGcOr*Yd`87ktb|
zEZcXA{_+GDM&>o2s=&ki_3S&4(Lw)t%d6e-r$0I4vj}|VapCuP=Xtz`PtDeoJNCH6
z5cB&keA*L!_VT{EZlRi3?jwHmq=){zi*J1Nvpm+iUr~c$KUnepYoGS=FBo^L_|HTesI^7r26+1Zg)2Xt-BrMmT{>QFzRzV7(GtIaq)I*o9NLbHBue
z|95z^5*=08eKtgelSe-erG#-ZU|XnsO-P5yWjv>%3iqXl;PnylWnX_d5nX{_faqL;
z*j)0J3G=mwifCVc$cT?Ph>*C5kw}S?Xo-Y)iO;1w@uB`P!nJwzj@;;tymy8^NP%A$j*g;VWhhuC=!-$b9{ZS64Ec)F
z^oy+6jenPr{MbMn7?1dvk*HUX{)dX>=Y@3PkimFqskJeX1cMcsj>003Dj9I#D3BJp
zjuM${%@dIn=|9Tll4=8a9C?o#NsTQDfz>#VKK@xcLOGK+8I&}*Oh06fG`W)oi4toV
zmGd_q+4zy#2u4?#F-a7VGFg=!DN9s|ls`z6O9_-iS(bftmhh97Ntu%U2sFE(KwtTg
zVab-!hk1CZV|q!FAk{Y)nSky1V|3FwS-FE>Xp=E%ltt;0Vo88x=!B7pmN?0iX9Y1-
z;v$4Rx1%sgNl7mY&%r$is$YDVk%*T&D<`At{-m_?l9Qn4oBzn|E5L=@YE8
znic4Cy@_y!*_Ln#m3av>)WVrj$(0B>n!{<9YuS&DS((zQmZj-@lv$mY36Z2DL%``(
z1j3xp>76oXm?NlmO
zmH#=SA3B-+LZC&uP2@tP7y3{&$)oqFqw`6h;+duUP^C%AinZvZd-;xk(WGG-W_~GZ
zeE_CrIzlO0j`nG#*qNhpI;RL)r`u?!SV}u(%BC7BGZGpmeA=f#*K<({p9dOrf;y;L
zdP(^QErX+(iK&>9TAPzfsWU46n?tCgIk^W7swhXfs0dYqBQclw(~F@BP0$Exj(T^f
zN>-2>nziMW^5A=)XrPrEo0p2Ix>}>+a+tfSlddXabmL#KDy(DWrpMY_F19z8>a5R-
ztI-Oqw>qu4%6`iFYO4AqE{d(B1*@vpt*pj4<07Rux}|zbr+%8Jy}GU#dZy^;rrXOpt`6{w(HLs-Uu^9`J7|X93tFJ5zcqO}KCo8i=g^N>)
zur+J4NXHc`39Z#?tv~*Irz;Dy0GqCLQXjmjvm@&qN~>FWh>1}viBemMRGW!aJGEJh
zwOf0&U2C;ptF>MmwqVP({zI#}o;w|T3wjJpVHJGf{kxsi)tg-f~VWVzc3uL+y792&5S
z>$sj9f|na$nM=B78@TfeKkoy5G&E2<&ux+vSZMti)(TfB%%x3&tsD)qeG`$*plzGf%A
zu#3FGOTE+EqyBP>z1u6J<6BSS3%_wBzw@h9Do>jm49J{rf#Z7#~T0D(a9KKO3#=#fFWz0)gT%0=0#9Z9C
zUi`&yjKe65#xQ)wb!<;$jK|}6$6y@7eyqPYe7@>i#4}9Dd)%gajL5yG$cqfbF6_o9
z49GTo{=I=L$bTHkPVC5_*~gna#+~fR865Pg_9LLOD!m5nPw=AH?EX$2d
z%_=g?*E}KC47|T=xWN3)mmJN?+{L)5&7VBV=d8i7tj=FwiYe67p>75
z&Cz7r(HQN~Aq~oYOk3)M1Uw4*k|R{nTrXfM<=@l%m#qO&DAKzRNt;
zH~rOME!c8B&3>K8$*kCX&Df1SVSSCWQmwge?bMci+2DMrwFlYMlg*u-yloBIk|NoQ
zN!fUM+TD!VzP#E{Ex@ATw_VvX2dJ=i_n+|OLtb$zYL4J5^F-R6wk*}W^W{nB(z-P4`f;{4Ed&D#E)&E44S
z-RX_q+s)qJY0SS(rSiR`QSGJlji|Y8-|t;{kImmF=->aXb?aTz&kfy49M=e*;D@c?
zKP}+CDcTQC)bTChw;kH?9Nyx6-sC;r9KPZE?co)^+zu|{6Ykz6zRdx?;5%*LhRxy)
zKGzWa;jfM279LVHZsYKV<0Ts5AkN#pJ+b&b*+OpG`VHPYe&ZC54e0g#>7lOaTy1-&Ug&l%>7Z`vjsEIbF6N%z>ajlGlP>FH
zzUD8U=VY$tJ0OR%KI5e>>W=Q}#7^b7UXB7S?2B&fKW^)!j_j~L?K|V2!JX;9ZsvH-
z>ulcbXdWO);v3^m?&WUo=Z@~_uI}s3?(OdG?rtj0s_L99vGb03+-mRnuJ8NK@BL2a
zO#`TX-soow?5CNu0k5~PB^vn7=Nn7#LxS-6$*T52ClE4>Q__^en2J71F%{3E)sD1J
z%kaZKrxCB=xn35-h*u`>CAvurzxp#U?=vB!oB3w&H=o}f{x2SCibGuEJR`Yi^4_r2
zj;TI>azMYbT*LB=VDdhpAc97Qx>&gZiy!&Pbz!6
z7Zfk^>Jb&Xm?UJMD>EG`L>@k9q4550_jix?d9U}t?Wr7(@3>XbM}w&nd4nOu`dqANQ5-G@QTr%3@%TU_XV>6=w1Ip3nJp
z?3HhO5jK(W>{eh%`1hUPTwoDJ>uC%r4fWCC4ZzY85h44MFU33;^&UZsnSVWqv+%3`
zEO-MCw;w#D6Z?;oG{|3w6r?}vx%3>|11?>NKxF!XHB&*hMi!uiPFsZiJNh`|B^P}U
z`?!CI?63Ka0=eSa42I}Hkx&1clNKpY0LUS48PmdpjY-;Oyw+t@mE{r1tHVriT15k5
zwb6BWvUa?v2mfQ7y!C~>3TsEK4Y4w`j
KcE91#0suQ2+}#ua
literal 0
HcmV?d00001
diff --git a/doxygen/img/Chunk_f5.gif b/doxygen/img/Chunk_f5.gif
new file mode 100644
index 0000000000000000000000000000000000000000..3b12174a479fb655614483e05d4798a12df6ef4f
GIT binary patch
literal 5455
zcmV-V6|m|@Nk%v~VSobQ0e}Di00030|Nkri0001h0^k7v0{)DTsmtvTqnxzbi?iOm
z`wxcVNS5Y_rs~SJ?hD8AOxN~}=Ptlk!1iE1|^00fM%S?q*RExF0|1@
zk=o#Tk5#emj83a#->1ghjKCFby7C1ZP|szo-K|nKTkh1`TG334JvMm
zj;-qxK4s9hjcM1cV8B%j2c~-{jax!;bP_=cr|F%+{!srAm6KR6q{xvadgVGfQ4U0a
z*hmqKDDBjzStz}Me1>z`ONk$K!czndR~Sn?5&D8Dlcdt6Os8DJlt-XSUr+I*5-0{1
z)qq@MDyn+1tE{0=XGJ|&G$p@D2&D>nx~Xc@xpeE=Wibfo-Kl%~`uz)dueHF03mZO+
zII-fzj2k|{Md;W|jFX+*vOPhZ4c9iPXtXsQ&4Li2%*|clh
zzKuJ#?%lk5`~D3)xbWe`iIdE_W-t!XMmC#ej<*r?=tm_vca!{0ajhbet5IHq`v+Ix
z$6a6HTDtmp)_rH^bzag)@bO`=n{J|Khx7FQ+u(=4+EDL(S082N0eE0LwYBgcG0urr
zAbS%{7a$OfG^9;)s+p$(gc91&poX+LVwHPw?e?8!622qHFGR3cP9PN_IF48!))>M~
zBrwNJY7C;sBZ=#!DA+YV8FkPdL9wNdfJ8*NQj{P#BSMH;ly$*vIl6M%Tp>bo(?#BS
zgkyCbW%y(;YQFf=2p#1(#fjp{Ij1fF?pNY}U!JgMNkiIopgWkSEg~rbQoQU+|blTDigwu@TJZMWGf
zdzZG}f;w)wH;6mzQs~C%Y_#ou`!2l82~&oRJ*;c%apn@6@4o)hq*g4eE@T^$bm5!j
zzX~sGBuAMZ*`crur(o#B77N<2Nx0@3ZgLx=*DR4Dr-5+7^2R~lO!$^e?Ztfd>vGKT
zs+%3lGT$7t%`fO|ZpAqNJhM_f>l!qOe==Kict^Y2G0<=h?Q~==DP3fWP)Ch5%~Us?
z^`cjIT~{w~Wx6ZKRg?U2#7jfE_Q@pMv&bzv{&OWf_u2guq6dTRHQ(Ldy&_I^zI-;b
zU;B;t4nNf$4W?>yowz!ONB*uamG$D7V3iU-n>NKEznwJPZO3NiQn;?7L4r{QL1TZKoG}{!Z_(lMR
z?T9=yqtnnxE_vb6j(pUc@jlqFKAv%jMtfifb67(L#!-$oOr-kg7)U`zZI1>EA0;J;
z$xF7$fpx^B%QTt3Li!Ga9Ms<;59vr8rc!G=jHD-V21<gA^_$;@a*!kL|Pp(~}yFlt^?gL!KtF&U{yWO_54
z-<%I5mx)bpX%n5)VP`tu>A42h@0sxY*E-kOO>&kKmHOvbkn5ZbSt#jsZVtZ)U&WCr$n`dQGZ%gq#iY>OWg=ll?qj+4z;ON
zjp|K%`qZo@RjX;yYE`|elc`$ut3@@H$6VRckGeCapc&~*mg-ArvURRb#9ytp63XMj
zl&*h^C^sj{KVt?qqk|f`?KLTRJq`lZF8ac++ki<
zqtS(~Mc*r0f&P}f<;5&=<@nwD=Ip$bo0@(RR@}1r2a=9ZX@#Sh$!2E^Ke5`(gLScf<<*BPTJOVPo8F#du?}iznifwy#8P!
z2b!}(CUlAA+hF=$cF$d$vu+t(XKryhf;qM`bMN~*U9OqUKpr%R(cEcHC)Uj{ZY`Cg
zeB?ADc9YzE4B7tyBG&SPxdyp%yiu3;kC!$J5F?KCz-BjTl{r+SR{3
zHLfRn=hiwI*@sbfnR#7iVsr1dKrkY=e@$Yck-F5>Ms%P1$?Qa%(b|#YLb{R7?sj4u
z+~GE~pk?T7VQ*XCV50T3%N>|_TldWX&+8BcF6~Me+%O0q_PlxQX%XLhcPiAj1wlP(
zWB+^M(Y9hD-;HBQHXLm3=D6%SZo!8SJmNyG_x;3(ae7ajDIiBi$xW`9{*hlCw$ePc
zd|2#BogY`>h53WCZT(K6J5l9lj`=2EZgG6`Tk1|%vA>V)mzv8=={q92Ov#ONiTk$6
zUSI6OxlVDQ0{he5w$%5gPIWc+n&!eyC787iYm~>G;Ev^c*nREs_t`z?CPzDWnLg~O
zBi`WlZoK2g&UO<680^hsywxU*cUbDZt~>`;mYb(!^8vl=bvFE~UUzfFs~zX2pI$du
z-ttEK{psATxZLM{XqR_K@TjMK?*Fb*M@N44U79uUAJO`k5`XXXg#7WLZr8W?zJqU?
zx$gbAOc%dB>daqz?BO^j%S)~E`xbxO^ZEQu!;7Vf8gb@>@5}f8?{cPvs;NV^ihr`|
zf2{lmtN-&KR#NqU`!|672Y>~LfCs361h{|($bb*%fDxF09ML_Sl~&`x$Afm5f^jlAVHk&L
zNDxc{9nS`cu;&JYXhUVVgGxAj*B6O;h=h}Pew27}q-Xwq9TkS~HYJ%8h;#Rei#Uat
zreTZ%2c>9=SchF0$cO5uiK8flviN(FSc}MWgvll)G}Vff(kF}8eeT7Izj$s?*L$Av
zg%@ayrV)%QIE=vriMEJ~W|)iFriMGmh(xoEaHu9(R*cPvhF7*x;V6!G_=6C3h=O8{
z;^=mwc!}9IjkI`;@o0_rh>z5WiTSpR3s#ni5EAI!Qde^Sc&!6f(wa_-DrB0wAP)(VV>pl)8G|T^km}?s{n&3rNRlfl
zhp3^5+W2XoHyhX-VCvi>KN;RSj_*=mheX(7p%21%Iec8unfm>0%skcBzu=a+wJmz5colL?qj
z*Eg<{l8qBAK{92HnM(sHm;hOFVUaox<%y!HK+iaiZCHY_BAMwhKWoXFffqlesgybS
zlR%l7T=|uHnUQUs-qHWm^5ml+Nq&J8l&{c
zp*HFjy4j<2MWQu2qf6?XqQaln38f*`q*LmmMEaz}xur--r7vovUOFlWikVd!r9Wz>
zILf1G8b&#~lx#?$=^3Qo$)#lqq&~=|59OwL8bZ>OIbX=9dKxmTsggx%sE2B&b?TmA
zI;6iDPY*&;fx4VRQk!U{6%9E4ff9(Rn+k!M>ZzQnshtX{pDL=Q8mgyCs;D}uoN87C
zVyVhuo01xEgL+rP5~2i}rH+cBExN0Wil}%oHp&&NvTAl{N~}S*rpIbRdrF>K+NEI%
zt&OUyb9$$5sH|<8tk=3g%&M�tv}t<_qEa%!O$8m>n=55wB6={T0_N=w}8u8xV8
zzWS`u`m0|$t>-GHy((_;8cp!(uU(d{0ef}q%B$s?q4j#NxtgzW3a)rcu!01z5xY7Q
zOR@bmu<=u&4r{ImORowGvWeQE7we;sNwQ#Mu_v2w^XjaN+Oqgcr~AsU4Lhy|tFlRy
zvNzi`Ijgf(nX%c~u^#^WvJ5M(H5;@>+paw;L_DjsK*O|6t9iyMvqC$xGaIw^y0BJT
zv>OYx2`RN*t5!A(w%hu%SNX71d$rQKwF;TAB722nyR>Bsw>vAhbL+ETtG0^jvT2*O
z`Z~27+qQ6Nw$;+ww|Yyow5zqHtFh#2t*?t`n9I9|>$|_J
zxUoyR=sLPEi@QL3x@;S{s4KjAN4(Fgy5B0j%7eIt8nUw6w`iNZ%Gnp$D`?;XIy^`Cn
z`&*cvi@Px`c8^6a}yX;%Q?>oS|>%bS&z#EJi9qhsCsK9(H!N~i++9@TsyzZ%b?#Yz>V9(j7h^QOCX*Is6jlvUbGX~YsAC5
zQggsK&bz}rY`TGK!YPcwK3r>A!BUrMs;TP5tJ=k2tg2y5#$#;8UtGp$e8ys|#-!S+
zXoFVmx5RflAa6xb>N>|}OF=J4HcC9f?Hj;RoWWO&#ZxTBI-JJ>Im3y}zc$Rs*W$=>
ze6|K${>ZsI$%Nd!Z@RjV3{i{B$)^FrpX`sG{KJ{7$-0ZlR-DQ!{KqyY%1BAdue@rT
zEXzF;$(2OKRUF8M48OWe!2q1DwXC@#EX)RV%f(!6$K08wJVB_O!?|3+yKKplhs?km
z%hhaO#Ei}Pd(G1f&A<%E;e5fXEY02!r+|GrL4*dJ+cAKlKt$^ya>?|J<0%G&U%c<-h9aB
z+{*}k(8x#84{gyQ{lFtl(wPy;CF;>1t-=?5%FfKtE
zm7CLpoy>yGXP5lUd)>^64b9
z9oL&}+VT0-sV&i!ZP53O*08_+Rt5dbq$TQZQW4~*|mM##f{yXE#3e8*56IN&@JAS9p3)R4ceov
zla8&{Yu(;wP1og3c)Km%ecjvkebw;I-TS@W{GHs9?S|xR)%m^9o~_^hP2f;{F6>I>P9BUv5fJgG5mo=Hd6*n&oa}+TkxPC+v;a
zUH;+IXXig+OjshXl@Ih{@|ce>OIXDl*A=AVh?|w1;L6};eqLV
z(h8ko=)GR6aE{>=Y7w-aCB&ZO$ByjDuI#@ZD2hJn`dx@C-i6SJ$2x|1#Z{Aby_wF=
z?XnydHt~YHl0~~tFrHu*3sIb8NrL3j$JJr(C_}`GRtZ8OHYRbPcf}pR-n7kbCKRDl
zX|9TJwT?z97EbjIuNRsAo_b{iko5BJF%pG2mQ$=7da7>ka>M5qAvj&W>qH#!fn-y?
zh!g;i@FIEe0YidhF^BE`y7zJ>2k%rEALvVRY*e!FM&VywL)A?dZ>w~e#Q)y!=fK%m
zsqJI??L;+Hp{~VpF{bZ+?_=@wZE_C{vGFA$@GuyQ^S%rk0rOK{@EyPGVK4S$k4*&t
F06RG`LxunV
literal 0
HcmV?d00001
diff --git a/doxygen/img/Chunk_f6.gif b/doxygen/img/Chunk_f6.gif
new file mode 100644
index 0000000000000000000000000000000000000000..616946db192346ca07db7e8a5d5e1858fec1967c
GIT binary patch
literal 4949
zcmV-b6RPY-Nk%v~VF3bR0e}Di00030|Nkri000010$~9F0{(=LsmtvTqnxzbi?iOm
z`wxcVXa(kprs~SJ?hD8AOxN~}=lag~{tpZahcg55h)gP%%%<}RjY@0Asr8D@YPa03
zw+Q%($K`98Fz)IBlJ+ork^c{oU3LK2H86UXGqVb*|p-p2YqxKi?Bie~%BAzt2Bp
z-w$x#zkmei4J?RI-NA$m$1Q9KQQJc;3QQyjZ~#Tc{)n&EFu)kbV#tgjNoFLuCnO-1
zDluwQ!IC2^BUNI8yy?=S&Yd>#Or+?trBIkTWp)fxv*^vFJcl`LnNko+r&KlmG#UtL
z&Z$~QQbq6*s-3G?w5Aw(Rw>q0U3CsMYt!q?K4j;7n!9!F%%D5N=55+nZPcuHtI8Fu
zH&ooOi~%bZI~KA}u3#%}Ox%p-$CEWXkNhk-PF>8TXNE3~y4vW}tTVTMZEZE|*_LF}
zzWvBH?%j$`^ZxBhH}K(feG@;f*f{d#nuIfd&c-?P>1Cu-zaC3D_U&1P@NFZ22^jBbl{|TsIgZS~M
zo(u*S)l-FrUD%g}8DdBwhZ=UcVTMz&HsOaRdWd3(Ag*|#iY-z&Vm%SY5MwSimUd$;
zIp%<%iz%}BEn!l1>iUq=rY{&?7KYu61Q_{XyuFl2AVBWtdTV
zn59uAiV5bJYNpAikY(y~pnJv52+^BtZ5iibU+#$}pKG>gXGBT<*=M1I9=c?pFjV=A
zq76Bk8J#a6=x3XTT8gNfa!Q0!rJi2;>7S6cz$h=HB6O-Jslp_w3xi?`YpAh;DpIQn
zy((+2v}TIos%P2y$)?;08ym4H0lWSJrMu3`YqPQX3ecI(>N@SREI~UAvaD*$3AX%r
zTN|E0;A(BT)iw)mKIERuY`fHg2yUsnq>B%|lf=udRb8_CZohrnOOLefnmaJR_;MKx
zzy=u`S-#mKTyVtx@>?r|6kjayt_SCtFvHO(?8wIBc1rNX8H?&MGy<5?9ACPUoB|j^35g18_`?Rtc4^8#UG^_H_q9jjU
zG|?{NEaTJCTzzNOq>>%Z(K$2ab=y@NEw*Y91gmG)Z+i_k4@>t(866XrBvju}?VWbu
zp3VJh-1n*-_T7$`+;pvmPyVhi<&}#fxa1OKyRX}J+nx6ggc}k$;Foi5x#Wolp4;Gl
zn{+`*9Y;-j-RyEPap`~Wo;m83Upx9byj#uz*Rhxzx(>lhHGJx(`wqS8n%_KPr?8i>
ze0Q{SIkfV<;MK40y-QF2^W>M#x8cy2a6ZA_QwhGc$@6HNzz{bf{_FZX-~99;m*|n}
zAx>XTuph??$T+>R?OjeHf>?4#tiJ_@8{6_ufReSqWi`+rM7h}fdLf$!8VD(#c?v5O
zN4G9i4~4>0pVJmFsSsAsE7vojaag!52yw4y)$3qdI8+$pnNJMpYnV99=BllotcO2r
z;);UkwhvNCH%hDu{$8G#tn$bVhT*vk6(kr#945|+6AVmGpm>V}iY|k5gi}+KbH9coAV={7f
z)!`-_w28oFj`ETtvmH3!2|;xdvY9k!B|LKZxodh*oKuu%A(EL5fKG;;Nc*NiK_yVR
zRWqOP#3n)h*XB>=Mbshmj08XD>CS`dlUWI!VMUc`PfXx*qZ-B3H0$|MGIVrxE5&F?
zkI>SYJ+yHQEorV|s=1W1)Hy4wsZXbMQm2tqrRIDCI=ea1m0XmXB>kx=llqNdHg&2T
zT`EeU`b(Ya0;fi`>I8zM&1OgSy47q1!8qh7PFF9B{b58Qq@mfJ?c4&s>r#X
zwO~Po>sq_`*Rwq{nuEP(UMM=Lnw`Ufl&gMysRJ>q
z!na;Fql1U;F)%FaeAdn(ZJbb#FmnoO>8?;`xMG*O0(4Hsb1Zp+aOuij|cEB@pM
zuebs8lCYq^95yj4IJ<*>?}MG(;6+RMD^y-{m)qQCNAnlbV7~OA%}Y~7&v(+Q4DFT+
z$LUa$nlI@6b9p^&%M_m))xnGNp($KrQR|w$vpuw=o$Klx!Wzq)PI2Mb{Aon1Rl$9>
zDQ9V1HfFC{wV+ljvSVFrw;H>Q*j6?(bv``g2=?UnIZYeT=A
z$chzpvr8*)c=L462fep?gFVJ}n;N{UeKXz6&F(=Xq&iECuM1o
z|2VL=26W&3cybkUuI7Wc_#q$f#K;@o>crOa&m-;j$OBT6ZcOA>*#7E}mwqvjCpsc0
zZ}-KQCEu%WOy*txcYv3&&hR?r`9|pV5k{})+YDvi``vrKA5QEc-Fm4Ni1mxZBhj(H
zYxWusQ|gbr)0XGY`@nu&|3M+@NnUy_bq{;hNB^p~9_YTw4)lcn)BDNxcKGL+5cbZh
z^Y~`yuxpx*DI(eZ0gr-_~&H#
zM@^&QfRILZ7`QqX=zj(kf|}-mABaQz2UOZ*Atwk{09bSy*m)d?M5!lHWRZfihI<6K
zebmQ*X;wcJmwuxKd@gi2n$d#yXK5`HAVXM#^VeWDXkz>`K11|zN4O|W!Fs3X7>jff
z{WE+3l!atiK3n*MV@G+6ByFvRhFW-ru@iuWK|{9p8DWT2UKEFBm=wYXh-W8=?2}&8
zMuXUwX;`>Fr35(GGYyl}EAyopF4u?SlRt`xiG(O;SJ(b6D6#7(H7Th<`YUG(?Am7-A$yKpVGY#E6UgV{mnNaKJN;
zukvu*Lqgb?hSnpD&WMgi_>Se6j7MjGve<)!ri1_Fgv$kmb0dX^NPAAFNIp1=N~JLW
z_+7}@h7o9j_6L!&u!2`r7cN*?FnA#{I9CV>e+ejA>XDK9$7MK}3=%1LZs>6UiEb%4
za}FtiFX;gOD2~xcRx1aAESOG16@<$six$~&;r{nKlsJ@dxQ~O^C`H+mNQjC?nFS(=
zbi3x1fF&K|SCw)_mD)yH$bgofCre`KZb8Xz
zYVRg6X{ZBsc4&di?QaJGYFcdIR{Q@lX<3)dbX6~)Q?ul
zleGzxoArK~8D@cLD&yswIp|SQxtEGZi)AUCewL7OIglTynuZvg%Q>3236{ZWnRdyK
zuQ^q$DVWdMfY)i5u8Ex+Se&^@k>rVQ{>Em2q_CFNNj1b-Xz)3mN+X*PXP&zmli8$c
z5~rN*NniGfoB9=+4Eb5Dp6R+=x0VlnDUok*geNM=4t
zYuuTm`Lv@yie)!yeIGhnK#EIrHl!%%F--QP@phqA3OfBco>%Cki^rW`>YbIxrN0$p
z(8!|%>S9X@m%OEhN5P7YW%rdB$qEt;ex<))^Qr)%1shk0&$3QoKRS``X0gqlfS
zx}FGnr~w*eje41>d8e=`Ymh4bp~|VKUCN#_Nv7!es2zHM2Rf-Kx2b}(pJh|2Xv(Pf
z*;b!gr^uf?Mb6|ma3>QpXK_3
zotkpaiWgvPwF$Oj@(!N)askv5JYda_Y6=WU?TNvN!^^I9Rp&`nG&Etr@$A
z8=JSGVzI5@uTZF_?ot&JF$;jQ>N>-s>`QkORliNx`)fPrs_URR!k2|lTjh}`eJquB5f>D=u)S)Fs)G|q
zRmW{s;j}9@N`+gpaYuSe$0`e4y9H~vWfZ}$XL<|TXgj!nRMa^{*NcdE!4-V6Ceuf8
z$8-sNd;8>~mW#EQd&0LX!5RE_t>?nQE4TIQGk&Lenzw#^>Af0@4bTQjJ&eOR7rvcl
z!iihJu7gVc^HNgVkr>;y)QX%SCBIPodz(bR)!W3^3rh`3yT6!KE~J0d<$*6ezfVZR
zIitN_dstv(xPk#>=?Hr$Om#BXOKl8jcZZ(mSd8}ANp)Oq>?l20I*!vt#xhEOn@g+m
zD2I}a{)RA&YKxqNiX5+yEVF+6y*o+C^*E2K*twcS$~m<;vpbufENq~QF6Kzcg6xXv
zgT~J@imqH>IxNQb8OpuvJaX7N0hh?Jqssir$|X3pxGKt%{L4d3Sf|5bn~ckW7sjK@
z%Y!`1&Ah(CYs)_NmVTpJ48x7J0ZN1j4tF1xNh3nC6H_{dB&`@*H`kc||D$p4Pa(z>H4GH+n*JOg#
zg6r3Dnwbl_Ww4trp<}BK!
zo!G`$+B?f2sV%gp%}K1Sv#Sl;vHjWbVcS!A*tZQGqpjO_-P^xyT)-{dL^a&SogK!F
z+-aNK%gvw6?c6K-+|kW}(oNkWRo&O^+Sq;Fysg~?!QI^*Ex8TenDO1?Ef7cIZQf3u
z-s=rl?Csu3^xpA3!SYStI8op6UEcSN4*9L$h6T@Ao3OA-&~W
zdt>6_$(%p;>@52F?!?#9&-}9H<6DHEJ`$NK$TsQlFPdzEv6t}-R+Wyh5Oo*HCdXKhbwTYre&OvV&4i5eax*wotcN
zRL8nl>$38V7vdtI8#gE;t)?7_F{g{CFIV@syYi_+_i~9#h7C0&c0S)S)aA-IR9QS!(Ft
znr7=iP^Gna7L%hTrv(D{3PMFt1+5~xqO77trc1Ap7hhu6mp)p0+1)BP7g`4;3hVjs
zsN`+yYTPNp{RLUAUzZr`B2;@~8N*^GeJIe{O|G*>=|p7HRdQBvSDw=EuB)|>H^lzY
z_;5g-;F75Mft&p!_gBh0P`%guidtRZz|=lXzdBnp9|3(a$UzUT=%?)AO34~LyR~$
zY=QWB!W8BdcbdY7lNmC2C4GI(thg5U<5JPyJY_{~e(M2Vf(L!qtg=X}oOomSThr_M
z)v8l1UUX~sSdR?PjQJT}89o`l87nk}wZAEUXYEZcZVld3UE#W#JTokaoDOKd
zx!s&=?dM#ynmJb8esO7A%<{p;&|G;Mi7}K!GB^b!>ArXN|8g)YcSY+S#%85LK#E9g
zZr%{bH?=C4hmig2BL}WI6C+FHv((EHhtjQm*TWgNY)Aff-ZfXIAjMCP`;_XElP!}rO<_xnAWeV)3Qs;kwB>7{o=oN=pFYmAzQ!FxoT=_$c)@idD3
zKO_}`#xv1RH(hbWE!mFL?d(R7fkI!aXL;cM%9LLfhVBMc^CpqMuoakF`H9*H&E;=aMlbf?NC!zaN
z!%a^P!GG{_T245Kb
z|A_ZD4WX}OtV`bKi!*
zT=lDHQOpkF&Ec20r6KAjb%m?$dCEB5I9`UYrpjF5UWv6Y+19>7tx#v!$qTEmd4*!l
z7bgxTYN~YJqN*S=!BbZzr0C9fU0F5d98
zSA55n$?q9p4|~v3g1e@%HHW+O47L8E#E>G%;QkiJ)8#by^0X;kx8`%R`8lgyaCd61
zibGZe$K3y{!B3g(!*m+QqgvayHU#hCCb3H9chYNv_eeI=i|0RLYP@4Fv{;L9(WDOE
zjp37~H}q>mi#@zodk1(2dXv4wq{+oe#?an1@TrywyFvqRnd|A1w6pazb^ae>8H3sp
zo~7Is*X)nuy%dYaDkJ(JSN7w=4(1;X0VFzVJ?46jxx-T_DOLIP`dQ67rG(39i&w7q
zqIazg3!q)-gfUoXyx2cj$fxIb4BVH~h8j3*e!G4S-gtvHr_vRFbkjDHsY%4HC+LJ-
zPI$us7=$!BG<#0HzrMw|H&gq8`ynpLR6xqACz^V@R^f+k(Wvan_z>GSqsoUtln
zb^m?2G&w$}ILDad$By?P3-v-{P?B)i-YrA|f~$^p!jqjIA5MHHICjG89~UL+7Wl>J
zt65|CRuf;X`b~R@XBhj^?j6KmWkU%tiTas_YWe21^d)}gUrH{|pZmP5q`j$=ZxLIU
z4;wjFW4|v`=i4B$Jd$j%YH(JKRh1F_kzqF*y&T7)Qd`1@lKF<@riR8deS*(bYZ7M>
z?ghUkh7rzl#;Mj&Ts*4tPn+<(B78Kt!*YZEo#}$)#6u^=1jP%A7Zno;iy|V)Usz_8
zZc=;~%=0(-3%hLnxWzeBSzMo?Zap)&-0<*fv!Xat8AcyiZdlW}ukf2rL-|$OEIdA6
zOY*?~O4nscvUvl)MZ~+%7V^z~M*g&~3}U`zpup3Tr07`~FOVi}vIxY1X_Tg2>Fd*j
zqHp%!=Q3jQf7VvH5%~XlbcRdoG<)|S@uICHudlz59S@WG>_x+Pe}ZCN^43HB)XSdVWCQ9ph>S{wyrtXQ5bN!FCagHAigYvH?D_
z{maUdO|yALq-a=HkBdhd)C^3(%a1E}h5DUbR}$o_YD?ghB+3OMdFqPrka?wvJ@E{fhT8t>LAwLOL|xtZ
zATn*nEoZ->ZVGi?@lbc3GVF9*c~7t|#@=v2VTNe544zl5DTZ&;a%K>w@y0Xl$$GE2
z3&Zwtr(YRb%#(f{xS)o9(wO9TIR1L@o;07>3ur4aGQ?`~PZu+>3n&O|-G>$lAH6_U
z+ZvmOHC^eJRJSPup1$*bS!Jm?Ezl*C%{wi@J!P}AV;*cR@C<*q?2bf*SLr)hf5L@VYW_`_wiM2
zhTlIW`y0?!o0Kl?$)?n2hRL+Dhi(!>wQ<{axW;g`jT#D#T1QIh9E$p5OKQ)FqMG16
zUYd<@7YwiSekt7)7eSxF>&}0fT*ib2&)GKrVeR0(x))|HPaO-kKO>*c^qP(!o@wtc
ze0L9;mOw*m+-^3fb)Wt@@(ZQ=LjATLz8o~YI(qB9ao=7JVrt+j8Wx}Jndot%Gx}Q6
z)lZ}og=@q!bbSG>>G{-9a|9~G^`Rq2<5$Va_GizjKe
zFGzQyz^l5z3)PM#+kQNJ+``_3&;{v7;N4=zez8D|C)^u1g?FNJ=#X=0vHP<>^m=ne
zYu18_Eej{w-tFhoO`*jW)FZX1lgi)|IT2>mc3yMe0q+Md6nzao0(#GF2Pc-c&8Xc4
zzyCq-n3=T$&bXMdhvB2~r0{ai&-_sKjo3*2ytVLkDsAu9oqumjK8XtN2w!<%7y24w
zaPT#}SHt@qfERLy2@ZSDe@iR#ec|H4yM5yNZguo<#j4;j@X_+UsPJ-QpZOurE3;b4
zZ@y%^c<^q?CG(e@?ZJ7VCasR9&xEhB_}QzDz2_@MzxSj#n#$2sj#2gZq39S@jtTJD
zS<|srIo2x2w!*QkaO?mbn{~%#ee^KqI9H9HW*z6Mk#n=-2<*69;kaAjxUuB82k5v5
zh~4dV+^$9UcpW#E95|?$%YuA?G&w#RkWJl^uUZ4K|45FUQA1CI
zY&rxSD`O~-ZjjBJA^N6`69o3@S$WD8={TUxp?Isc-AgYRY$rN51yr718+ml1<+Yk6h-32qXLE}CcNs-MH6c%farshJf
z*KmKQ}AD~mCNmyp1UHtunu931X53)
zxFWDJy!5jLHIx6^5LeRB-&jM8P3T%nUS4h);Bwa9E~sB#OOdWxS?A^?5H|ps)ie!%
zw9Qx_#|FsAa~+~5l(>)4NQwEeY?0>F%7Da&4ZF$L#{*aye33NtBk9}*g7&)6Z9BjW
zk^8hMA5Rhm(7QK>kqH`U5AbJwgrBtE|AHo?1us$4Dsj9mZ%QXJ&F?#Xg0oLr&NS&b
z8)L;{!?|kRW27Jgbw8WMap237QySK7$_7c;<%P5Ilo4+L$
zPwQB2La1HW>CGE5%j2OPHP?JY7HyWYdXih+cmU?*noqSh+ej7f*k(fo06OiEpYMO%
zAvH@FwUe2ER3-xk5@e?tLahqF-W_(90k4~lvgu5b|0z2m(4sr)#ZJ>+f864(KI0N|
zfg=6)Wp`-HzO#kd7)Xauh1}3I|JeLVZ&CeOq?bz!Qqpe1WO8BoqkXxhl3`T?Fj8b#
zWr}|Ii6M2uFXMkKJwA5m-@r-kFKuoOFH!od{eUafHw2za@(Pl5xWp9+dXt?$rL|fH
z3ppzI=>R?7Iy?YGoop!iLuStw0?Sr?xOcY3{Lex
z)$r{t`JStjJ2ll9{@0FHGx{R?;B+tDr%1Gf1k&qI@-)y;ZAHBP`#sX6%tPgym59U%
zU+{=3|9(M-__-*t<(oCG59|BL3d+sdt!_la9|(W
zqEN`_e-XGKu4CXT!fu_tR+r*cLwbLf|In@FtA_w)xOr!1aTNVgbdO)J?BVi`+>fxKd$21L-^hGuP
z3Qo9G(lqe7$RheVwkM#lJ()ATJ$Iv_>oqjguYES$^W53ckwTCl?z(8RnO$9qSH&wr?zH9XlL=t
zV1W)g0mv)19!8uMG#f51j=Mm2@knhsvzUTbIsV_6*wj`x{VGfSmV7*Fc8rv*YCMxF
zsa!-{8^%+r%UuWkiXsbhD4L3Ocj$$!|2q5QzdCmJ-)0f_Cp;)%FDdWc<(0*Zuw=FM
z&jb1u<5~3Y>PnArVB!_BsY%M+{SrVIIc!slvYtQZa~LnY4J#h~(baJ(fXx&w=P?EX
znK6R{$5s<|CRlVFQhE1h+`ECN1Xp8jy_LI;%NWR0N}5x=+PtZT_CgF0-L>sFxl@%f2}h-vP7n2)f>4
zpLj8PyXZ0XS{Zc3zMrv;*yIr+avPfCFzOX!t~$dfj+?smD8S9%ZH|ios7S6nvY1B|
z#p)lZKP3my+NW7>E--6YBw9i}h3#W%S7S>sY$=1tROp&$!o_im>phG&R+r4*7psxx
zzoYz=tXgBj6QC6_5UGO)`;;{MJc=kt#eQ4wAf$-yW?~d$3ofn2Mpuc`dB9)K?fr9G
zRWxmu1Wj^A<{yRp;YSV5M9sl23dhs}q#^w#MM|j{2M9duAag~G!GebQWaF)-Y2>Vr
z%4^qm34KJ_#QyWRuDqMZN>`qD4mudbD|38FEz;zJy=;A%5<^hYhB+6Y{5r7MI*F@d)*Ob$)7XL!!Mt3|aR&}($-{*kKT~=WXauC_ZAb%Ur?m1y(^sdP?VZFnLrVS#*
zHWR#bEkblGz}mAJwI|-`A;4}HBdNP~Vd>XHsyKd+p9c0gJfC2Ov9kfurLM))_VL%n
z!O>3bitEnjp}X>iA4Q17168hMLi#-fIg-y*7xvM5L-ARX1+dwn^=y-zTUgxjT~{sH
z(zn0+Qf@uZ&^3;x-5hjywrdDgHp(Bn2JgRP{*Zb1W6wICVb4OEHYL#Phg*e)#D7Vp
zK7B?Q;d5#K{TV_0+TSrpMAoG*6xWdo6HU;qlztkSe}-SW{tL$ImPXhUbgTB+78Y@q
z#H?s)$T(v3ewj?$@*~$HtalB6@G7@X5{(!xY^R#3yFM)-J8nF5{Xm5>9Ya)O^h;we
zTn{Xby)d#$0-Qy4E8RpN@WVEA_G2A7Z2(V>z(+WFrboaAjdcCEzg}Zk0CRb8
zTx*Q(_6IvsfP*xv!rG)Wzc;3#0uTdYGAWooaK8b^&?I%`0MFQ)>`SF-@26iMKPglk
zlK7||!TCjh=9hH;VobW@Q@<>95Y8B{4WBJ6+m$*wzo`91bOe{qKCQ1uX#3z2ST+8;
z-HjYJ?J5>jiO_$nN=JJh-s74V_F4fWv*ySIMO!42CmY#*94Lihqd*Pq@fpgU*e>r1
zw&Tgqg5h>M$)iK{o9t4lFVRd;70DI5`c9Vma)RN=hcTq7CLH{QB%v-SOsGfGtctZ79sGa}H=T#G`JA}s7UvbJKVcx6
zS#`djYT}wuZG*o!V2_xYfqx9}urS2*fd9MHAuISvx#Q
znfz4fi&T2$3!Xyl*rbQB$E4GjoL_igMh0O2eNtxj6V=yl@=OpFNCvig>K
zbGcpwZaIVfNKLX|E^@_PEl$f>#aTAKlCZi~;PbpZK&8+{EZJy9>3Ygn$#rYN>j
zX+_nAgME{Quu=h?N;CMuJqHc|kjqeG3<k8zewd)0xMxV1JF$8GMz~Y6b`t9aT)sF82
z2hR?V9}v*uRDu*StXIeHrfFqMcx`5dZvd77_C4(uUbI7LGrn9WOu7SD>I$I50GR~p
zb6YN#Iov}z@jarz0>MeO-`JTh!uBcdY1&i|-;!E(^yrM_oR=jOyd+;1@|MLKUilQ;%(h}N+1cK_|Y
zZ67Z1ZJ$qUe1JfR9Qt6Bu3#cQU-oEn1XZvf!v8{EMkW?G7o4aqtJS?hY80SkyDM!L
zfO&zMln7dAN#Cvc<@lw3LB2>brNq5&tXQp>)Sd?nA5blpa*QDR6msm#clw#T^29P?
z4G81f8qd61J8+~&@#STAoik;EZUQgXIM}Nnq5OPm(g28M`{8r#ekmE_72B{SaG2{;>_Bb0L+ngJb<>dq~p7_;lKV0d{0@tPo(z4dS8g#w?z3e
zik43Q)f~5f;C`T*2Doq7;ppd`&F@U%RkOsQJ8CAcfylCWI*vd%)^|++-mgGasANo3?r^QFZ{59H*F^
zk2Fi)P8HhRd!S8|k9~z~m!>Hn?o>%I+(cAC!0Cd7xf)Ws$Am0
zC8L$AJ(vRez$&78EXs24S7e6O>n+;1e;0m^@xJj}$o?A!s(fc^c1(umOyOkLxE3t7
zYt@Lq?s(A2>O5v6Vh{NsxK%4cA#e66ZRhasHL{Q!`HH)1y>jAwg)^@QLf36)9WabE
zCFG?mY+Re{6EfXSg%w%vo6cVjB4x~pt%?@TGVFAzaHZTlneT5Rjr3FmlL9fTR2RFp
z$6x{c7F@hXwX3KmD3IVKPkG#`mvp#8g~^>BhHK6n5ZRNPM(dc<5Xc|8Z|vD+{QerH
zQNRi-JMQy_&%5$Hw4&I)o1Ov6oVd3A+A3F&oXtG%F7c=wlfs2@>-z@rjm~X}NIrGy
zbJ~3`@`XD11IS4q6$Rev`xaQuFxuLgi*O7N%>iej_7$FC_gs&|Ga3#0SPU|Miq`~_
zmaX%8FukE4;^QayQzV#VV5n0V1}aSHL*nE;lQN44to
z7=IuWys8M=9m06@8b+WS4(tQrQp{4?WjCpF0XC|n2YXmKb=!cCbPM&PrI52~!{3Fz
z?x9)4_|228sWPFd$!-cOol3^^G(dV4c`e2)Ti)rteUF;dmq=TL;KQ@AC~Q7{>;vrk
z6^XS30=$m+6;bUdztEdi9X2%xO3Pa(nQ?;2e*thR@BeExDFIUCq0q-t#v)
zKNN@;lOF8}$nw|j_$x=XKer39r~RPUqVX1W<<;!?NZ}T2_me9BXmSBq7OP#@f+&SlqglR|FC~W1@m?
z|9XgZpFsw$ZO#)
z(bEGqOBV0dSwV|>`e>B{M-WQ{?yc08a%~v~rXAe7xA#x&bzudmX|iVYvkF|*_eHN)Gm?7N
z0y#>IrBA=J4>6eX$Z73vZ&7vMk00n?VW>F8Tbv5u>_V{P`Uj)|8H@duEeJYq)8_s(
zO49bI_Opmf7T$u3GY+^9_LcwytOvBWx39|YotV{PRaWQE@74qqZenek
zrTS$mEE8pypW>YDkThc}rbl&vM8BJF-?jbrdK-y)(-kXNf1(BuG)}Gdb>$2P166YZHN;D79KftY{8Gss+6r%^S@aD
zrxJCvDtZmf&Mk*0x)j`d8-&Gf!1P_>-#;$`A2A3IjmmxDl&mIK(Guun)$G$s*H|ZN
z`>G`ha%eHb{ZSTR(*;7q+wL|d{Ldm~ilnkaSm7gV+?PhL$7U<&!O|DV)I9!?RcMy3
zj*v+$EKe0{?iBl;(YE}`WjMR`
zW|deUoCZGr$n1zg#9`)!fnF42IgN22AVZXomxli`soz%&&p8E0LhPqx0C%K%{yMBCm%HRd2HA?j2?Ry*qFXfd#rD
zF`p4z%m{B`xoP|}~R2ROK*-(MFUK_Ow`}6?dIMbEBuZVVt&<;14Dm&7U
z4uaA6jn&ie4_c7_8nXt6ap;5nm(%F$z)Tv*>pRnRXyU?*dXm-bH>JYWX(owY>HH8T
zV$($gHrohaeX(dg+K%;q6Icc|>32Vjs&l(ptYojC_nqS0gK`R75%{wkkhr`FEyz$i
z;Afby{q0M&z4Hq=K(*TH$jDfx*G4VGo3L5>zk#uRYJC1XtS~1w0;`d8s!;@p`o+ii
zGp07L@?zYfNChqZppj;<&u7rpv4y2|JQZwB)`{ZlDGL2zP&BX75y}t)&hAcbat&Bu
zh3-5C0CO=FrXRAQ(p3@0WWUuk88Qd?UNB=azD>7b$4svMD7>ge{
zb7+;|;DllbHXZB}hSZyM3U;U%*|ZsUwt>1BJ6Te0KDiICcNhJ~O&eBTs5Lgi-2ENG
z(;LW^PICRkPDD~OaByt)p6IheD|xZtW&TTFmIW4=l}TgITA`tS@#0OL9w4lw$^{#V
zRDl>M>k_O+&+cIMTrcpy8k>V28*iIV;jhTILi48y%D=^zw6FSvx8%5AlHr@M4Y3il
zqB@DrW@ZmFRAT1QHok_n8nzcLN>oBbZmFR|<9j_Wz_BJ&&4H;~Jr%Z61fe277mCcdUk2e0EgzOa%Ln-@};f
z>humi2+lGUPPp@Bacss=R{q%5}Y(2K*Xp>ha`l;Is4f$-}3#FMb9W*#?@bPc6#1NmJ=^3<&Kn`O?Lq&aAN
z1_y@(p#S|v|EBHr61$Rq{^09tlsA2BhoOAFWvI&&4kPh#*E99eSC
zSkh_x`5I^T?)e?#kSm*;+XnwMioiJTN4@48qDk@u7I{hR1=P<+3sJQ=!-ncmK`>M|
z`(Q-VJSRADkXaOKs|Zw#TP&jFV8zV5_7}FwBS}z*g`HNMPepVxwlPF2#_%+8U=3D5
z%x#MGASLvNUlq)Cegr;L%ecrH?hVti!oqT59`qh}n-bAaTXu`Puifcu`hZkPvKmv?
zet-FQxu9p-hGjcaaWn2wrFQ~x-OnBK-46qr05cd{KBaBl1ec=mof@fZHEHOM`!UVH
zW$eyZ)CAI14ODg5LSBFE$?6Tnf%{HR-ukKSJqB^4O7-G
zJbdU
z`F@dIIQSA~7Y30rs(OOH##j_m+MbF$x$r`x+2`M`9SM%G+l3b%fUOtUg~wDPyD%2`
zd*+8We6gY^@zh+K?-#QRqcTWf7q->FSq#@>*Rxk0d(ZiiJH&2|rgAiuV^sZpC^|+J
zR_Wd`0Xo(y$67`G*|Dvd9^=?nICg-J&AMZ=?l@PW8{&?0mE#EPxLZ-{xLbihRL6~_
z=n>Gyabw9D?zo|Iv*RA1<94l+<94m%&bs3vh~ps$S|1(HtI$I($DMV@o%Ls*M>=jS
zIc_XDZY()&EIG0jTWv`WM-0XhgK^wgdLk$0xUuxV?8efJi7PBg#DRO8a|4(|4Vaq=
zGy0_50KkNK+ZO`1`yX8u*W(%OT^XUNnN|Tq;MV3%2VD%ungsGPf!lLVeTeDsTX@YC
zAEK+)Ld-h8L{m9YITita0JDA{7!*}yeiy8*Q^sFGkSBs;6=SnzCtz`DY7yYOxhUYA
z0O9WCCSW}Qkz&?6{R&WWz86ipYe6q^7(Si5j=~IS<|s}FI)haW9qA`4vQ@^53n`#7XS(S;Al@&HhJ6Z4bq{LBf
z2~>lqYj`_Ke@2nN7%_BqHWx?)+4VMk4q%cV5kkGt7|JcjGR-QM9qDCbkz5RS4I=ku
zHLkeK>@H~zcdhl;9K>ZKXWcp^LR0!qJG!{-FMyZ>$@Bs9ybY1pSof*zf~)e>72(S}
zPj9{m!qHPY=Vr*C8>Y`~#}Tj18()EX!O@Dzn8eL|>P`8ouYvHG&}n%0bcCWpd~6j}
zV1+a+fqxz$R!&Y$Ncm-|aMivdcE!DGHX4rK#CYwe6ZcP8J0ELg=VaKxE4$IimJ>fDm5>MBT|Q@PHPJ62}xwlk``X
zGtEB{3LY4h9FjHeCzQZ$7l^d7SAC|RUx9zodwyEP6s`NCGXE(6g8ko>NpPe2bt}t!|7LV8HdpT~W^*RLhRh%6QdI6#^tp^QZZ7xORAGdIo
z@!J&NM9~ghgp_u5?y8v)8q~hd1nJm@Jo0dN*BlV#nJbWf`RJmrsBIm_e}Sh8iV=jE
z9*Q?$hz)>QpMgdj$9ev>Nw>}YmBWB+xXGswg^KxWUPve!Z?pM2w@Hm(?H>PSKBjKw
ztvO3lB5NLaHf^F{8xD%usP!umW>!p!=D6GO<&iKpYGC+#gf_wO=n)6ysKjdoIe03S
zy|%E`&F4X5rzT{Bcb)S%U=_}quHQBPK339OnYg5uN&Lgl*l|jk*#nx!bG&mPamJBq
zSbIUJ=F8q^Ctsm$GvmzLFWZ_{I~!$xn7w+IsqyKC_8o034aV=p6b@pwIS}Dht*IMz@=8rSA2=ERPl64#&tnG
z-m1lmcvNYHcpArW-Prs-bz|oU{&?)&8P&klfh`e(pRNBT;vB{7Al@8CC>7;--Sd=j
zx^d9Be>1e%35x##N)?IQP1i^OH|-~|!2+UFv|
zFdFzqU+eA!#O$r;$U1E}kU~zL$6}FNuNSH`2bF%xY##*XWijNwj8KNPz3E7>izv$m
z;W7N*FpZ5xH5BM4TgLzmZ6|DO2spyu!TvdFWPdycIv6+Un_X}t7l|@nH=g3Gt7$y7
z*ha+5hHN5vCGiG9L_cZtV6{xN8ydl&Dti6Hyh@v$@p8mY^>xc}z1QF;y%${8x8Aw>
zO(%GkttXMKSQO!2YuztOY@=9WUxZF%sC-3|4K_zZJJgdvG+Qo$vUC!yjU{r%nSXjt
z{RtB5)^K!sj9J1UQ;f625DXUPFP^1e3$OM!iYng_hp5s+7=Ab_Ae0@(6Mx1{!r915
zDg^eY-TbLGMz?~mThXAlsMPh82kPNA_4+t)DJ%5jRaRQhL8XSJA49tbbHQ|VD`v}W
zT1%z_Q|$az%=8C=d>_Sp_9q!!i=1grYv7yWtl<`^f@u*!S5AIO(%0^*>p$>U5?OM+
za1%|FgwJ+jGp%atu(`!Zw$w5+Rfumas%^qn;eo2FRqGakPq8qP{Dto|_{MfoJ$!!v
zY-E#}07@2#qOhVXX$UV=CEFB~fLxFLTYEc;r|XWP-PQoAsGz@hCa`F;SjPLK>D7B}!-gZc1#U^|u!{U$lbT>hoG9~qBnpKQ@>o_kMi
zXm|RAhFr>VT#N%F0tC(BxuLE}d4#aBp-LMF1?^-m76)>C^1W1vSl^|8XPRqwsdKdr
zzX@FpzmVv@j7=C8igED~;qE#DSi`w_L}xq!mlpAyJIWs{?uu^J^mrKt>I33$ZNA6?
zb*-+v+ICgR$E$tIHDj0Z^ShAZ7S)FdfX#Q(<(1Pgr=g^N)tEkCrp{N#ZFqf>EbBDQ
z+D9e$-WLo%)pa^}Z?RaTQpKeBByfL*rU$0g$eV^_!^!?Ho8BDU#JZ^Z>+`B;gnbg3
zkvbOBz-r(OH*wZ;Kx6M%PyCF9_`d0sS
zPh!+y73=)r*UfDM2h$cEQUx8`%R!KcZolZ+4j`nD5cio3Mqm%RLQ3rqg`p(_+`YV7Cf%OeuRrEOOm
z@@nMTuZJV8a^j80Ra@xR?y(+N1&E65in59pnJ&FXUVOrc<&cgm0B_8R6*eR1RXKhOppu078(z!eGl
z5MB)o{DyP%;fLJehx>ak*KF`gY^uG8_JZi#^CNd2Yi|!S?DjW_7_pX(FO#p2rxmCfH7Hz(
zx6ksVEV9o>FZY3swnp!jRzo@E-(p4Qi4NXhroae|z}60?e{Khm-x94~eIHoX{z?NH
zyhf)#VO>l_f?FriLEJdA#PUc|Z=J1s+nKstkrApOmshdGw4dp+Bd#0J8Mqz^Z)fmk
zobxL0zuSe(y1mIbu5loSMy+$c(2@-cT;#ss*rr`N)(^CaINq^UhR+VusOy(ivc!nR
zzQ$^Neg&#;&m)!as*&;qs7jJyC(^DyTMG&fj!SI|EoFE>9Ds_K^`GLHJF5L*
ziqW>*bJ3HCKO0SatH~I6K|9&|uB%zIcRaUz@S#z#WOrw9h`&c#SP#qgsajlN1WaaR
z_+)WfQ1T*B=n1^IGSphjzTjTpKohOPT}FN>2R_mz5rPKcR_NkOJ0
zk|D{DC|&ia)(h21s5j&oCW&ZEe8jK7VKni)r2nFCUDO182+Z}e6d4Q3gqhF()eMAr
zlWWujo+0K-TEk2DNXldtb|G*~eH_zw^=k$h<0IM;FUG!Cne}dbR3c64G}5sx?i{R1
zppe)<)?yl&g`XqaYcV}jxhGSesVv4XmiQlwq#(;Gg`l9Ie)-)f$Ge-qt1x8V#+VHbBmmi+C3HF~6EN1ZcZ
zne4DJYn5U83|{94WPj=8L!5=P8)w74o~Rp1c(c?`CELce1Gfh`&ksRa9RHdX7;wvv
z%ZWV9(cBHhJyAqJb1Z;9Fb0at6se=EsLTY-b`bc^bkW^o#JR1Q^hDphA*(=GtLd0q
zrczr)BfVbig8Q|PN!Qvf>1g9u)$`%|_Fo;p-7=>Y<)i}J~T;X`r1#qbxmlxP8Py{+U5+==b3>o0W9g~A&
z?#1H8Sh&_PIoQ2{j`b83k#I~7j>+MF@8r;Dy{r$b(BJRO@ec`ZJ^t6e?~?zm-dz0A
z8QsP+YnuL?8usXD^Bea!ygkMJZn5g8oC&WVyYbqa8$S8ey-4~)#hd?nZS0~qPpz9S
zJv{z{xyQDgEn!|Av|2B+Z!5DNniSc}t6THBgFoewH=H?B)uwC`ZRO~oNK4F|*X_Te
zw{)l0Z`m*%pe}n{8Ox)MZaV&62q46FbocsHIT=KS
zINlVM#Bp3+^kkvQ#nF*=!I*Ol89OrOm>e9F162D$)?3Hq;8;%|7ov0Q9R6?EnB@mA
z(4*4A%X@w;xQG&b7Ri8)W2IScBWnUHjMiwp=7d`Z4@Nw5)A{DkhEG<{TEhI2qAFHy
zz-$@%);Q2a#Q$}kb};s>mHY}%
zEOEKVC!=^_0?Hopdv|$dk-jb+>dsS6S!qiyg~ym46^bs8g(qs7D&ONlajh}B!Ugb{
zzXCnMTGC3Dr%=8=fTCOF@I)^wyrk=*@$WUV+n%m~i!^+ZTU@mO->2{3T9_u*KV{bIGYuZRRxE80>zNH`0(4cH#VHE^&LdvhaVqBTHQ!*
zRI+?&U(TUJWmAJtaBX){yQWREoEodgh|czwDIB_*Kl
zzIaTzbevblhUHy9b+spZu}M@p(+&Skf=i;|m3L_-7yR~VuF~T;aR}6i#vaMN4!7khy*TUzrfPN&xU(n`+K+-LUX|RuLFb40#D=jyJ^$>cjUv |