fix(chunked prefill): don't schedule prefill if freeing kv cache

[toslunar]

FIX #5578

This PR makes the priority "running > swapped > waiting" strict. The previous code has chance to schedule new prefills even if preemption/swap happens or there's a swapped request, which (unintentionally) changes the order of requests.

#3853 says

1. Schedule all decodes
2. Schedule all chunked prefills (running prefills)
3. Schedule swapped
4. Schedule new prefills

and it has been implemented that (3.) is skipped if (1.) or (2.) fails to schedule all. The new change is to skip (4.) if any previous step fails to schedule all.

Re-adding preempted requests to chunked prefill may be OK but it seems better to implement "partial preemption" of a sequence (i.e. freeing some tail part of KV cache), which is out of scope of the PR.

---

The previous PR is #5633. The commits there are squashed into the first commit here.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

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 either: Add ready label to the PR or enable auto-merge.

🚀

[comaniac]

LGTM. Only a style comment.

[comaniac]

btw can you add a unit test?