Speed up restraint centroid distance calculation

[andrrizzi]

Possible speed enhancements:

• Add an optional keyword to Reporter.read_sampler_states(, atom_indices=None) to read only a subset of positions from disk.
• Check if it's worth implementing directly an algorithm to compute the centroid distance without going through mdtraj.Trajectory.image_molecules().

[jchodera]

Another possibility is to convert all restraints to use CustomCVForce and retrieve all registered CVs for each CustomCVForce and store them each iteration. This will add almost no overhead during simulation and make analysis near instantaneous, even if the cutoffs or bound state definitions (in terms of CVs) are changed.

[jchodera]

This should be fixed once we convert to CustomCVForce.

[andrrizzi]

For record-keeping: This was implemented in #1059, but it hasn't been merged to master yet. It's now in the multirestraint branch, and it will require an openmmtools release (of the features implemented in choderalab/openmmtools#365, now in its multi-restraint branch) and some fixes to the analysis before being usable.