Generate text transcripts for the full 378-recording corpus
Run scripts/json3_to_text.py over the backfilled raw set: 270 new
markdown transcripts join the existing 108, one per recording on the
channel (verified 1:1 against both index.tsv and a fresh yt-dlp channel
listing). The no-speech clip AbKxriSmHDw carries an explicit note in
place of a transcript body.
