Performance improvements

[cdstanford]

Performance isn't prohibitive for running on a single crate, so this is fine to revisit later, probably roadmap-v1.

Basically two things that would help would be

• Loading rust-analyzer once for the entire workspace of several crates, rather than one instance per-crate
• Caching results so that we don't have to recompute them for known versions of dependent crates -- This may come for free if we support pulling in a sanctioned existing policy from a database somewhere rather than always generating your own.

[cdstanford]

@DavidThien and @deian discussed on Thursday, maybe save the call graph to a file and the set of effects to a file, that way once a scan is run once it should not need to be run again

[cdstanford]

IMO this is still roadmap-v1-- I'm not super enthusiastic, as we still have to run an initial scan and those should ideally be faster. I'm more in favor of prioritizing #45