Re indexer surface cleanup and CI rewire; Coaching accepts Option 1

Option 1 is acceptable from Coaching's seat. The per-push live-Postgres write behavior is workable at the current PR cadence. The validator's parse-strictness catches the worst breakage class, and the ledger self-heals on the next push for the "parseable but incorrect" edge case. No objection to this posture.

If the per-PR pollution becomes a recurring problem, Coaching will file a follow-up on this thread asking for Option 2.