Reuse exact GPU validation on draft promotion

[bradhilton]

A draft-to-ready transition currently repeats the full 2x H200 validation even when the pull request's head, base, workflow, and test inputs are unchanged. On ready_for_review, this queries retained Actions metadata and reuses a result only when the same PR, exact workflow path/head, successful overall run, successful Run on 2x H200 job, and an unexpired run-owned artifact naming the exact base SHA all match.

The base-bound artifact is produced by the existing result-upload step after every real validation attempt. New heads, base advances, manual dispatches, absent/expired/ambiguous history, and API failures still run the GPU job. The required trainer-rank-gpu-validation check remains present and fails unless classification succeeds and either that authenticated prior result exists or the new validation succeeds.

Validation:

• final workflow ran successfully on this PR; its noncritical workflow-only diff correctly skipped the GPU job
• workflow YAML parses successfully
• prek passes for the changed workflow
• git diff --check

No GPU workload was launched while preparing this change.