Faster-whisper

Improved thread handling and a host of other improvements as well.

Added "distil" variants of the Whisper models for increased speed and efficiency.

Added multithreading so the GUI should no longer freeze when transcribing nor when simply downloading a model.

**Full Changelog**: https://github.com/BBC-Esq/ctranslate2-faster-whisper-transcriber/compare/v3.1.1...v3.2.0

Patch release to make setting cuda-related paths more robust.

Improvements
* Modified setup.py to utilize the powerful uv package to speedup installation by approximately 2-4x.
* Removed all Ctranslate2 quantization options except float32, bfloat16, and float16. Nobody really used them...
* Upgraded certain library versions.

Using sounddevice instead of pyaudio. Upgraded faster-whisper version.

CUDA is no longer required to be installed. This program will automatically download the appropriate .dll files to use.