Use content hash for mlinter review deduplication (#47830)

* Use content hash for mlinter review deduplication

Instead of skipping whenever any review with the marker exists, embed a
short SHA-256 hash of the findings in the marker and compare it on each
run. This allows re-posting when findings change across commits while
still skipping identical findings.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* Store mlinter state in PR description for O(1) deduplication

Instead of paginating all PR reviews to find the marker, embed a state
block in the PR description containing the findings hash, commit SHA,
and a link to the review. On each run, read the PR description (1 API
call), compare the stored hash against current findings, and skip if
unchanged. Update the state block after posting a new review.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* ruff

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>