Add persistent caching for execution concept-set resolution

[egillax]

CachedConceptSetResolver currently caches only within a single execution context.

It may be worth supporting a persistent cache layer for concept-set expansion so repeated executions can
avoid recomputing the same concept closures.

[azimov]

I think this would be a significant time saver for a lot of projects. My proposal here would be to use concept set checksums as the core identifier, as the same concept set can have multiple ids within different cohorts, so there will need to be some mapping computed at cohort runtime. This could create a race condition issue with simultaneous generation though