New package structure

[jchodera]

I'd like to refactor and reorganize the package structure to be more sensible. This doesn't need to be done immediately, but the sooner the better.

Here's what I propose:

• perses.analysis can be retained
• perses.annihilation should all be migrated to openmmtools.alchemy
• perses.bias contains various biasing engines that would be tailored to different applications; I'd rather split the concrete instances up into perses.apps, and just retain the base BiasEngine class in perses.bias
• perses.dispersed.feptasks contains several things:
  • ExternalNonequilibriumSwitchingMove and NonequilibriumSwitchingMove which should be moved to openmmtools
  • multiple tasklets that could be used in various protocols, so should perhaps go in perses.protocols
• perses.dispersed.relative_setup contains application code for relative free energy calculations that should be moved to perses.apps.relative
• perses.samplers.thermodynamics is redundant with openmmtools.states and should be removed in favor of using the openmmtools classes
• perses.samplers.samplers contains a number of perses-specific samplers which might eventually be migrated to openmmtools once we figure out how to generalize cases where systems can sample over multiple chemical species
• perses.rjmc contains a number of reversible-jump tools that should stay there until we can figure out how to generalize cases where systems sample over multiple chemical species, but SystemGenerator should be migrated to openmmtools
• perses.storage is an experimental storage layer for the samplers and should stay wherever the samplers are

[hannahbrucemacdonald]

A lot of this has been addressed, but we can revisit the status of this in our meetings next week