Investigate migrating away from file-system-cache

[JReinhold]

file-system-cache has some problems:

1. It's a pretty big dependency for just reading and writing to the file system in a specific directory (we don't even use it to specify the directory, we use find-cache-dir for that, which is also sort of dubious).
2. It includes all of ramda (lodash-like lib) because it imports it with a wildcard.
3. We've had some typing issues with it before, where it was distributed without the necessary sub-type packages.
4. The versioning/releasing is funky, as currently the versions 3.0.0-alpha.0-8 have been released after 3.0.0, which doesn't make any sense. Furthermore, 3.0.0-alpha.8 is currently the version on the latest tag: https://www.npmjs.com/package/file-system-cache?activeTab=versions

Problem 2-4 could probably be fixed upstream if we wanted instead.

It provides a nice and simple get/set API around reading/writing to a directory. It might do more, but that is all we use it for, and to me it seems like something we could easily do internally without the problems listed above. As far as I can see it doesn't do anything special to improve the perf, it just uses node:fs and fs-extra under the hood.