Cache pre-render strategy — enqueue a render job for every article that does NOT already have a done job. Idempotent: re- running with only_missing=True (default) only touches articles still in the gap. Use only_missing=False to force a re-render of everything (e.g. after a voice-quality bump).

Pairs with the auto-enqueue already in the intake worker, which handles every newly-ingested article. This endpoint is the one-shot backfill for what's already in the DB.

Renders flow through the standard provider chain (kokoro → openai_tts → elevenlabs → cartesia). Failures land in /admin/queue with last_error set; retry is operator-driven via the existing retry-by-error-pattern endpoint.