WebVTT output fails after transcription despite being selectable

问题描述 | Problem Description

Problem

The transcription UI exposes both VTT and All output options, but WebVTT export is not implemented end to end.

When either option is selected, TranscriptThread eventually asks ASRData.save() to write a .vtt file. The save dispatcher rejects that suffix, and ASRData.to_vtt() raises NotImplementedError. Because export happens after recognition, users only see the failure after the expensive transcription step has completed.

The CLI has the same capability mismatch: it can read .vtt subtitles, but commands that directly export subtitles cannot select VTT as their output format.

Steps to reproduce

1. Open the transcription screen.
2. Select VTT (or All) as the output format.
3. Transcribe any audio or video.
4. Wait for recognition to finish and the export step to begin.

Actual behavior

The VTT file is not created and the task fails during export.

Expected behavior

A valid WEBVTT file should be written with millisecond timestamps. Bilingual output should honor the same four layout modes as SRT, and selecting All should successfully create every advertised format.

Version

Reproduced on upstream master at 95842ec.

日志信息（可选）| Logs (Optional)