Sync plain-text transcripts to audio

[wassgha]

Summary

Adds Descript-style sync for plain-text transcripts (no timestamps), including paste. Timed SRT/VTT/JSON import is unchanged.

How it works

Same idea as Descript’s “Use your own transcript”:

1. Import transcript → paste text (or choose a file)
2. Optional speaker labels: Speaker Name: dialogue
3. Drop media (or Replace mid-session)
4. Whisper runs for timing only (plain text); timed captions skip Whisper
5. Your words are aligned onto those timestamps (Needleman–Wunsch; gaps interpolate)
6. Reference spelling + speaker labels are kept

UI

• Upload source menu → Import opens a paste dialog (“Use transcript” + “Choose file…”)
• Editor transcript panel → Import opens the same dialog (“Sync transcript” when replacing)

Test plan

• Import → paste Alice: … / Bob: … → trigger shows pasted transcript / syncs to audio
• Drop media → progress says syncing → transcript uses pasted text with sensible timings
• Paste or choose an SRT/VTT/JSON → still skips Whisper and loads immediately
• Mid-session Import → paste → Sync replaces/aligns against the current transcript
• Choose file… from the dialog still works
• npx tsx scripts/parse-transcript-test.ts passes