Fix: Add Windows (AMD64) GPU support for PyTorch installation

[andi1479]

Problem:

Windows systems with NVIDIA GPUs are currently installing CPU-only PyTorch instead of CUDA-enabled PyTorch, making GPU acceleration unavailable on Windows.
Root cause: Windows reports platform_machine == 'AMD64' instead of 'x86_64', causing the existing marker conditions to fail.

Changes:

More explicit Linux condition: platform_machine == 'x86_64' and sys_platform == 'linux' (instead of sys_platform != 'darwin')
New Windows condition: platform_machine == 'AMD64' and sys_platform == 'win32' → routes Windows to CUDA
Fixed fallback condition: platform_machine != 'x86_64' and platform_machine != 'AMD64' → excludes both x86_64 variants

Key improvements:

1. More explicit Linux condition: sys_platform == 'linux' instead of != 'darwin' to avoid ambiguity
2. New Windows CUDA condition: platform_machine == 'AMD64' and sys_platform == 'win32'
3. Fixed fallback: Excludes both 'x86_64' and 'AMD64' to prevent Windows from falling through to CPU

The same changes apply to torchaudio sources.

Testing:

Tested on Windows 11

uv run python -c "import sys, platform; print(f'Platform: {sys.platform}'); print(f'Machine: {platform.machine()}')"

Platform: win32
Machine: AMD64

Before:
uv run python -c "import torch; print(f'Torch version: {torch.__version__}'); print(f'CUDA available: {torch.cuda.is_available()}')"

Torch version: 2.8.0+cpu
CUDA available: False

After:
uv run python -c "import torch; print(f'Torch version: {torch.__version__}'); print(f'CUDA available: {torch.cuda.is_available()}')"

Torch version: 2.8.0+cu128
CUDA available: True

Validation: Successfully transcribed a 40-minute demo video using GPU acceleration.

Info: I didn't know if you want me to update the uv.lock file as well or just recompile it urself I can do that any time!
Thank you very much for your great work!