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