[CI] Fix attention backend test for Triton

[tdoublep]

Purpose

This is a retry of #26663. That PR is showing some really strange behaviour in CI (CPU tests running things they shouldn't). I wanted to make a clean PR to double check there is not something screwed up with that build and that it is indeed reproducible.

Test Plan

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[gemini-code-assist]

Code Review

This pull request fixes a test for the Triton attention backend by correctly transposing the KV cache to the format it expects. The change also refactors the logic to separate the cache transformation common to both FlashInfer and Triton from the logic specific to FlashInfer. The changes are correct and improve the test's correctness and structure. I have no further comments.

[tdoublep]

Have triggered the "V1 Test others (CPU)" job here: https://buildkite.com/vllm/fastcheck/builds/44832#0199e7a9-2267-4771-a13a-4e4b853f3159

[tdoublep]

Closing since it shows the same behaviour as the other PR. Will comment more on my findings there.