Transcripts: new and existing episodes are transcribed in the background (published transcript first, else whisper), with a status line in Health

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

README.md

@@ -116,3 +116,7 @@ The lyrics screen shows a **Transcript** for podcast episodes and audiobooks. If
 straight away. Otherwise a **Transcribe on this server** button makes one with speech-to-text: the bundled `whisper` service in `docker-compose.yml` (faster-whisper behind an
 OpenAI-compatible API) is reached at `TRANSCRIBE_API_URL` (default `http://127.0.0.1:8000`), model `TRANSCRIBE_MODEL` (default `Systran/faster-whisper-small`, downloaded on first use).
 CPU only: about a quarter of the running time or more. One job at a time; results are kept in `/data/transcripts`. Remove the service and the variable to switch the button off.
+
+In the background the server also keeps transcripts up to date by itself: every ten minutes it looks at the newest `TRANSCRIBE_KEEP` (default 5) episodes of every show in Podgrab, uses the show's
+published transcript when there is one, and otherwise transcribes the episode (one at a time, at most two new ones per round, a failing one retried a few times). `TRANSCRIBE_AUTO=0` switches this off;
+playing an episode or pressing the button still works. Settings -> Health shows how many are saved and what it is working on.