Get Eval Run Evidence

Fetch the frozen customer-quote pool one eval run graded against: every quote with its id, tier, provenance and whether it reached the graded prompt, plus when the pool was frozen. Read when drafting an edit to re-grade against the SAME evidence (then pass evidence_from_run on the next run so the score delta is the edit, not a retrieval change), or when auditing what a verdict was scored on. evidence is null when the run never recorded a pool.