[llvm] workflows/premerge: Generate a ccache artifact for each pull request (PR #124311)

[Aiden Grossman via llvm-commits]

boomanaiden154 wrote:

> You can implement this using the cache, but it's limited to 10 GB (for the whole repo), so that's not really enough to cache all the PRs. By using an artifact instead of the cache, you can ensure that the cache for each PR will persist long enough to actually be reused when the PR is updated.

Ah. Completely missed the fact that this purposefully used an artifact. I thought caches saved on the `pull_request` target were saved in the context of the fork (thus counting against a different 10GB limit)? Couldn't find that wording in the documentation at a quick glance though, so I might be forgetting something. That would make the simpler approach a bit more advantageous.

We're also running the CI on PRs as if they're merged into `main`, so I think a cache from a recent `main` would have a much higher hit rate than a PR-specific cache from even a day ago.

https://github.com/llvm/llvm-project/pull/124311