fix(stt): re-anchor whisper's word boundaries on the audio, so transcript trims cut where the words are

fix(stt): re-anchor whisper's word boundaries on the audio, so transcript trims cut where the words are #137