Openlrc

This update introduce new features.

New features:

- Resume from previous translation.
- Add atomic translation for src-trans inconsistency.

Other Changes:

- Update default whisper model to `whisper-large-v3`.

This update improves preprocess efficiency and minor changes.

Other Changes:

- Introduce multiprocessing for loudness normalization.
- Fix the .srt generation issue for video input.
- Add preprocess options, which users can tune.

This update addresses minor issues.

Other Changes:

- Split audio during noise suppression to avoid out-of-memory.
- Improve translation prompt.

This update adds a preprocessor to enhance input audio (loudness normalization & noise suppression).

New features:

- Loudness Normalization from [ffmpeg-normalize](https://github.com/slhck/ffmpeg-normalize)
- Noise Suppression from [DeepFilterNet](https://github.com/Rikorose/DeepFilterNet)

Other Changes:

- Now all the intermediate files are saved in `./path/to/audio/preprocess`.

This update switch the underlying transcription model from `whisperx` to `faster-whisper`, which enable VAD parameter
tuning.

New features:

- Switch whisperx back to faster-whisper for VAD parameter tuning.

Other Changes:

- Update translation prompt
from https://github.com/machinewrapped/gpt-subtrans/commit/82bd2ca0d868f209d0e0c5f7c04255523daabe3c.
- Change the default parameters of `faster-whisper` for consistent transcription.

Emergent bugfix release.