Python Autodub v2.2.1

We finally added support for running Python Autodub without an NVIDIA GPU! But before you get too excited, let's talk about expectations.

The Good News:

• The entire pipeline (Demucs, WhisperX, and F5-TTS) is no longer hardcoded to cuda. If you don't have an NVIDIA card, the system will gracefully fall back to processing everything on your CPU instead of instantly crashing.

• Linux AMD Users: You win today. If you manually install the PyTorch ROCm wheels, this code will automatically hijack it, treat it like an Nvidia card, and run at blazing fast speeds.

The Bad News (Windows/Mac/Intel GPU users):

• You are officially on the CPU fallback path. AI transformer models are incredibly memory-bandwidth starved. Without a GPU's VRAM, your processing speed will drop off a cliff.

• You will think this program is a crypto miner. You will think it is stuck calculating pi to ten trillion digits. No, it is not. You just REALLY should not run this on the CPU unless your idea of fun is watching paint dry or grass grow. Expect a 10-minute video to take several hours to process.

Developer Note: I am officially marking the CPU mode as Untested. Why? Because I am not a masochist and I refuse to sit here for 4 hours to watch a test file finish rendering.

It should work mathematically perfectly. But if it breaks for you on hour 3, please send a log file so I can fix it without having to run it myself!