Transitional: legacy schema-1 export metadata consistency + admission regeneration unblocks release path

[ElbertePlinio]

Follow-ups from PR #14 cap-round review (all transitional-state concerns, none can certify a release because the new gates reject schema-1 inputs):

• Regenerate all 10 admission reports at evaluator schema 2 from clean HEAD (this is QA integrity fixes: prompt-safety, diff_size, reference audits, clean admissions (Phase 0) #5 item 5 — it unblocks the release-report path that the new exact-version gate intentionally blocks today)
• Schema-1 anchor exports serialize current defaultEvaluatorWeights (missing the legacy diff_size 0.3) — internally inconsistent metadata on a rejected-by-gates path; emit legacy weights for schema-1 anchors or drop weights from legacy exports
• bestObserved exports over mixed schema-1/schema-2 runs stamp schema 2 — stamp per-run schema or the minimum schema across selected rows

Review refs: #14 (review) (round 5). All three become moot for fresh databases once admission regen (#5) and the v1.0 run (#8) land; fix only if legacy DBs remain in use. Refs #5 #10

[ElbertePlinio]

Admission-regeneration items resolved by #22: all 10 reports regenerated clean (digest-stamped, gitDirty=false, sandbox-enforced, evaluator schema 2) and the release path is unblocked with digest gates. Remaining here: the two legacy-export metadata consistency items (moot for fresh databases).