Add capability to cache fitted Stan models [WIP]

[jburos]

In some cohorts we end up fitting long-running Stan models as part of the analysis process.

It would be convenient to be able to cache these fitted objects (in much the same way that other data points are cached, ie variant-effects, etc), but retain the option of invalidating the cache if the underlying data have changed. Perhaps something along the lines of a memoization would be useful.

This idea isn't 100% fleshed out yet but putting it here to keep track of it.

[jburos]

Most of this work has been done, and implemented in the stancache library. Next step is to incorporate those capabilities into Cohorts.