Add concept of `samples` to `Protocol.create`?

[dotsdl]

Proposed Protocols like NonEquilibriumSwitching take advantage of equilibrium MD followed by nonequilibrium MD, where the nonequilibrium MD uses sampled configurations from the equilibrium MD as starting points. Under the current gufe data model, a Protocol such as this can be written, but repeated executions of Transformations that use that Protocol would perform equilibrium MD many times redundantly, wasting resources.

To address this, we could add a keyword called samples to Protocol.create that works similarly to extends, but takes as input ProtocolDAGResults from certain Protocol subclasses. This could be used by the consuming Protocol as a source of sampling for whatever it needs to do itself.

• in the case of NonEquilibriumSwitching, we would ideally want to run a set of equilibrium ProtocolDAGs that then provide sampling sources for the nonequilibrium MD; the NoneEquilibriumSwitching protocol would then use these inputs as sampling instead of running its own equilibrium MD
• more generally would allow for efficient re-use of existing sampling for Protocols that can make use of it
• could make it so that samples takes a list of ProtocolDAGResults as an input; would allow the consuming Protocol to decide how it handles these collectively
• in alchemiscale, this would translate to a samples kwarg on Task creation, similar to extends, but that takes a list of Task ScopedKeys
  • the Task wouldn't be executable until all its samples and extends Tasks, if given, are complete

[IAlibay]

Interesting, there's a few things we should consider here but it would be worth discussing!

[IAlibay]

Maybe one initial question here is "could this also be solved by passing an explicit component through? maybe a type of serialised PDB?"