Multidimensional Repex and AMD about brokenyank

[jchodera]

repex.py currently supports arbitrary multidimensional replica exchange. Because it uses the Gibbs sampling exchange framework, the "topology" of replica connectivity is irrelevant. One just needs to build the appropriate ThermodynamicState array for the desired topology.

Accelerated MD implemented as a CustomIntegrator sounds like a hack. Since it's just a modification to the potential, it seems that the best way to implement this in a consistent way would be to request a feature in OpenMM where the potential energy is modified by transformation through a custom function that can transform each force group separately if desired. We can submit this as an issue to the OpenMM GitHub to see what Peter Eastman thinks.

The ReplicaExchange class is fully general, and could simulate any desired combination of alchemical, AMD, and temperature dreamed up by the user. The convenience subclasses ParallelTempering and HamiltonianExchange simply automate the construction of ThermodynamicState objects. While we could provide some convenience subclasses for mixtures of these, there isn't a straightforward enough protocol for how to mix these, so I'd rather leave this to the user for now (as an active topic of research).