--dump-segments: stop paying for the transcription twice
--segments could already skip ASR entirely, but nothing could write the file it
reads. Now the CLI can:
lyric-align song.wav lyrics.txt --separate --dump-segments segs.json -o out.lrc
# fix a typo in lyrics.txt
lyric-align --segments segs.json lyrics.txt -o out.lrc # seconds, not minutesThe dump lands before alignment, so a run that dies in the cheap half still
leaves the expensive half (Demucs + Whisper) on disk.
Measured on the public-domain demo, tiny model, no separation: 7.02 s for the
transcribing run, 0.04 s re-running off the dump, byte-identical output. With
medium and --separate, the first number is minutes.
Also adds Segment.to_dict / Word.to_dict for the same round trip from Python.
Raised by the second half of #27, tracked in #28, implemented in #29.