chore: Cache random stuff

[I-Al-Istannen]

The CI spends quite some time building the cargo project for the initial stages. The directories added in this PR are marked with the CACHE_DIR flag and therefore cargo should be able to deal with them becoming out of date. I just want to see what, if anything, this can save first. There are a few considerations with cache size one has to think about, as we no longer really have a stable hash to base the cache on. Maybe we could build something from the more internal build files, but that is a problem for later. In the meantime it will just bloat the caches, generating a new one every CI run, and then hoping that GitHub ages them out nicely.

[ZuseZ4]

@wsmoses Do you know if the assumption that the oldest caches get cleaned up holds for the OpenStack runner too?

[wsmoses]

Don't know but we can find out. LGTM but can you name the cache rust-xyz just in case caches are per device or org

[I-Al-Istannen]

Yea, the names are pretty much random atm. I first wanted to see whether it helps at all. The Cargo.lock file in src/bootstrap also looks mighty tempting as a hash cache key, that might reduce the amount of caches generated. It doesn't quite seem to be exhaustive though (I have a yoke crate locally in my build folder that I can not find in the dependency tree). But maybe close enough, if the automatic rotation doesn't really like lots of caches.

[ZuseZ4]

Going from 18-30 to 3 minutes is an awesome speedup.
Thanks a lot!