Improve exception handling for protein mutations

Recently we have noticed that handling nonbonded exceptions have been getting a little too slow for protein mutations, probably due to the fact that now are alchemical unit can be pretty big (since it's the whole protein). Maybe there's a better way to deal with this, but for now we are just iterating over all the bonds/atoms and handling the exceptions on all of them.

To do that we are just manually getting the forces exceptions and adding/removing/rescaling(1-4s) accordingly. I wonder if instead of doing this manually we could rely more on the createExceptionsFromBonds() method for the NonBondedForce to improve the performance when handling exceptions.

Note that this probably won't improve things much for protein-ligand systems.

Originally posted by @ijpulidos in #105

Start by locating the code that manually gets NonBondedForce exceptions and adds, removes, or rescales 1-4 interactions for protein mutations. Compare that path with OpenMM's NonBondedForce.createExceptionsFromBonds() documentation, focusing on whole-protein alchemical units. Done means exception handling remains correct and is measurably faster for protein mutations; protein-ligand gains are not expected.