Change GB from single-point energies to one-sided EXP estimates of free energy

[jchodera]

I'm finally in the process of changing from using single-point energies to one-sided EXP estimates of free energy differences for implicit solvent hydration free energies.

@kyleabeauchamp: You can monitor progress in this repo.

I'm currently working directly in here: https://github.com/choderalab/gbff/tree/master/examples/gbsa

We can migrate the code to open-forcefield-group once we get things cleaned up.

[jchodera]

I'll probably divide this into two phases:

• Generate trajectory and store uncorrelated snapshots in memory or a NetCDF file
• Compute estimate of free energy change from stored snapshots

This way, we can trivially add infrequent reweighting updates if we want.