[V1][Metrics] Add GPU prefix cache hit rate % gauge

[comaniac]

Part of #10582

cc @markmc @robertgshaw2-redhat

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[markmc]

Thanks Cody!

[comaniac]

After the second thought. I feel it might be easier to simply maintain the cache hit rate of the most recent N (e.g., 1k) requests. This aligns the logging flow better because we take a snapchat of the scheduler stat in every step. It also makes more sense to me because the cache hit rate doesn't associate with time but only requests. WDYT?

[comaniac]

Gentle ping @markmc