CI disk space issues

[Nemo157]

https://github.com/rust-lang/docs.rs/actions/runs/6799427301 just failed with an out of disk space error.

Looking at the restore cache step of https://github.com/rust-lang/docs.rs/actions/runs/6799427301/job/18486983082 it's restoring a 3GB cache, which is both slow and probably contributing to the disk space issues. (It appears to fail while attempting to create the new cache to upload).

[Nemo157]

(The cache also appears to be only partially effective, there were ~280 dependencies built for a PR that doesn't touch the lockfile).

[Nemo157]

Running with debug logging, I see quite a few duplicate -sys crates included in the cache, the cache action only removes files after a week of no access, we may need to use something more aggressive about cleaning it up.

[syphar]

It seems like this is a systemic issue with Swatinem/rust-cache#158 , and not sure how to solve it.

we probably could use cargo sweep and delete unused files directly or a day after they aren't used any more.

Also, when it's coming from all the -sys crates, it's related to the amount of dependency updates we do, which means we also could purge the cache once, and then see if it happens more often.

[Nemo157]

Since we are doing infrequent mass dependency updates, I think what we could do is not restore during updates, use that as a sync point to purge the cache and start a new one. I'll take a look at the action a bit more tomorrow and see if that's something that could easily be supported by it. (Preferably while still keeping the incremental lockfile update support for PRs that only affect a few dependencies, so somehow keyed off just the mass updates).

[Nemo157]

I can't see any reasonable way to do it from just our workflow config, opened Swatinem/rust-cache#177