Transformers now runs llama.cpp quants

Hugging Face's `transformers` library now natively runs GGUF models, the popular quantized format from the `llama.cpp` project, making local AI more accessible. To achieve high performance on devices like Apple Silicon Macs, this integration reuses the underlying `ggml` kernels for efficient computation. Users can now load these memory-optimized models using familiar APIs like `from_pretrained` and run them on their personal laptops. Furthermore, the `transformers serve` command allows you to easily host a GGUF model as a local, OpenAI-compatible API endpoint.