Every segment row in the database, across all transcripts.
Returns segments sorted by order ascending.
Bulk replace: deletes all segments for the record, then inserts the new set — in a single transaction, so a concurrent read never sees a partial mix and a mid-replace crash leaves the old set intact.
Segments keyed by their primary key, indexed by
transcript_record_id.put(neveradd) so re-transcribe overwrites cleanly.