Distribution checks/comparisons

[JosephCottam]

Single distribution boundary checks.
Two distribution divergence checks.
Visualizations to support both of those.

This PR takes the work originally done for PR #39, bases it off of a more recent main branch and renames elements to better reflect current thinking of its place in the workflow. (Will close PR #39 without merge as this PR subsumes all that is important in that PR.)

[djinnome]

 this looks pretty good except we are no longer using the datacube.nc serialization.  The current output from sample_petri  is aPetriSolution,  defined here:

https://github.com/ciemss/pyciemss/blob/b15d214c188d84d02bc64793679c88fea2030da1/src/pyciemss/PetriNetODE/interfaces.py#LL20C1-L20C40

[djinnome]

@SamWitty should we serialize the PetriSolution to an xarray data cube?

[djinnome]

Also, I am not sure it makes sense to check distributions of state variables at a single time point. It makes far more sense to check distributions of trajectories over state variables. Is there a handy visualization that can accommodate that requirement?

[djinnome]

Also, we currently do not have a visualization of the ELBO, which helps to see whether a variational inference has converged.

Good convergence

Not so good convergence

[djinnome]

WTF convergence

[JosephCottam]

Deferring the ELBO work because that is not normally returned from our tools right now (maybe in the future).

Issue #164 details next steps for distribution comparison, interventions and calibration.