chunker: sparsify if all-zero data was read

[ThomasWaldmann]

related to #14:

once we have better sparse handling inside borg (communicating data vs. hole status between chunker, hashing, ...), it makes sense to add a sparsify functionality.

assume we read all-zero chunks from a file:

1. could be a hole (no space allocated on disk) [we can avoid this with upcoming change from sparse map / file map support for fixed size chunker #5561]
2. could be file space allocated, but never used (fallocate)
3. could be zeros stored on disk

for the last 2 cases, we could add a --sparsify to borg create:

just compare the chunk content to an all-zero bytearray.

if it compares equal and --sparsify was given, the chunker shall (at least) yield "this is a hole of length X".

later it could yield (if we can detect somehow):

• "this is a not-allocated hole of length X" (case 1)
• "this is fallocated unused space of length X" (case 2)
• "this is on-disk all-zero space of length X" (case 3)

[ThomasWaldmann]

this is implemented by #5620 within the buzhash and fixed chunker.

currently, the chunker detects zeros and yields as described above.

the hasher maintains a LRU cache of (hashalgo, size_of_allzero_chunk) -> hashvalue to avoid computing hashes of all-zero chunks again and again.

note that this does not affect storage (yet?) because in any case, borg will compress and store the zeros as if they were normal data. so the archived item does not know what was a sparse hole or fallocated unused space or just normally on-disk zeros.