Speeding up the HTF

[IAlibay]

The HybridTopologyFactory is now a lot faster thanks to @ijpulidos' recent changes #61

However we still often face cases where it may take up to a minute to create a hybrid system.

There are likely similar performance improvements to thoses added in #61 and things like using more set operations which could speed things up a lot.

[ijpulidos]

From profiling we have seen that most of the time is spent in the .atoms method from the openmm topology object. I don't know if there's a lot of room for optimization there, unless we find out ways to not having to call the .atoms method as much. In the _create_hybrid_topology method it's only called once per topology, as far as I can tell.

[ijpulidos]

I wonder if there's room for optimization on the OpenMM side of things

[ijpulidos]

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.

[ijpulidos]

Or maybe we can get away with no exceptions :) https://arxiv.org/abs/2504.14398