feat: Add MPS support for Apple Silicon devices

[nogeonwoo]

This PR adds support for Apple Silicon (MPS), enabling GPU acceleration for users on modern Macs.
When trying to run 4-bit quantized GGUF models like calcuis/higgs-gguf on an Apple Silicon Mac, the only option was to use the CPU, which was quite slow. This change allows the model to utilize the MPS backend for a significant performance improvement, making the project much more usable for Mac users.

• Modified the device detection logic to include a check for torch.backends.mps.is_available().
• The device selection priority is now: CUDA -> MPS -> CPU.