Create or replace a data entry's live transcript by fetching a Recall
async transcript JSON from download_url, mapping participant names
to speaker labels, and writing the result in a single commit.
Broadcasts the change to connected clients. A transcript with no
usable speech writes a "no speech detected" transcript and returns
200 with written: 0, so media can still be attached. Returns 400
if the URL is unsafe, unreachable, or the payload cannot be read.
Returns 409 if the transcript is already wrapped in an audio/video
block (media has been attached).