Add support for the quantized llama 3.2 models

The new Llama3.2 models have just been uploaded to HF :
https://huggingface.co/models?other=arxiv:2405.16406

models showing much faster inference and smaller size. Would be great to add them

Thanks

The issue names no files, tests, or entry points. Start by reviewing the quantized Llama 3.2 models in the linked Hugging Face listing and locating Candle's existing model-support entry points; done means Candle supports inference for these models with their smaller size and faster inference.

Written by the indexing model from the issue text.