Add MPI parallelism and rotatable grids to GIST

Version 6.22.0.

This PR is mainly a big enhancement to the GIST code (@fwaibl @tkurtzman @EricChen521), with a few other fixes/improvements thrown in.

Addition of MPI parallelism to GIST. GIST now benefits from across-trajectory parallelism (benefits the solvation energy calculation). The post-trajectory-processing entropy calculation is also MPI-parallelized to prevent it from becoming too much of a bottleneck for long trajectories.
Adds a new keyword to GIST, rmsfit, which allows the grid itself to be centered/rotated around a group of atoms of interest (rather than having to rms-fit the whole system to a grid).
The order calculation in GIST now respects imaging. To prevent this from becoming too slow, a pairlist is used. The final speed is comparable to the old non-imaged order calculation.

Other improvements:

Fixed the variable type when summing in MPI for double precision grids (was previously float by mistake). Shouldn't have affected anything previously since I think nothing MPI was using double precision grids (only single).
Add --ndiff flag to cpptraj to mimic the ndiff awk script used for numerical comparisons of tests. The awk script was getting too slow for large files (such as those output by GIST). The cpptraj ndiff is orders of magnitude faster. @dacase if there are any places in the Amber test suite where you think this could be useful let me know.

Amber-MD / cpptraj