Listener-driven render-on-demand.

Resolves the article by slug. If a done render-job already exists, returns 200 + the audio URL immediately (idempotent). Else enqueues a fresh job with priority="high" and returns 202 with a Location header pointing at the public audio polling endpoint /v1/articles/{slug}/audio.

Concurrency: at most MAX_IN_FLIGHT_RENDERS_PER_LISTENER queued/ running render-jobs may exist at once for articles the listener has forwarded (intake_listener_email match). Excess requests get a 429 with a Retry-After so the client backs off — matches the contract of the per-IP rate-limit but uses worker depth as the signal.