Apply Curation Endpoint

Flush saved curation decisions from the primary cluster into the data cluster's transcripts schema.

This is the in-app equivalent of the offline load_to_postgres --apply-curation loader. The dashboard records decisions on the primary cluster only; this endpoint runs the same idempotent apply logic against a write-capable data-cluster connection (DATA_DB_ADMIN_*), re-pointing claims for promoted/merged/renamed/ rejected topic leaves, applying per-claim corrections, and writing curated disclosure profiles. dry_run=true previews the changes and rolls back without stamping anything, so an admin can see what will move before committing. Idempotent: re-running only touches decisions changed since their last apply.

Runs in a worker thread because apply_curation uses blocking psycopg connections, and is imported lazily so the data-processing loader stays out of the app's startup import graph.