Add option to cache integer data sets to disk

[drroe]

This PR is aimed at reducing data set memory usage in certain cases. The first data set to be used this way for testing is DataSet_integer, for use with the hbond/lifetime commands (addresses a user issue on Amber mailing list). Adds new command:

usediskcache {on|off}

When usediskcache on is specified, whenever an integer data set is allocated by the master DataSetList, it is cached to disk via a NetCDF file instead. This uses less memory at the expense of speed - seems the code is about 1-2x orders of magnitude slower when caching to disk. This will be a work in progress while I try to improve the speed somewhat.

A test is added for cached data sets.

This PR also adds a new keyword aimed at reducing memory usage in another way: for 1D standard data file reads you can use onlycols to only read certain columns from the file:

readdata <file> onlycols <range>

For example:

readdata myfile.dat onlycols 1,3-5 index 1

Will read columns 1, 3, 4, and 5, using column 1 as the index column. Note that if index is used in conjuction with onlycols, the index column must be one of the columns specified.

[swails]

[ci-skip] only works for Jenkins on the amber repository, just FYI (and it has a - after the ci)

[drroe]

[ci-skip] only works for Jenkins on the amber repository, just FYI (and it has a - after the ci)

I think it works for Travis as well (see here). At least it seems to be working in this case.

[swails]

I stand corrected. :) Thanks

[hainm]

I think this PR breaks pytraj build ( error in dataset integer).

https://travis-ci.org/Amber-MD/pytraj/jobs/505631275