Update KVCache implementation

Matches ml-explore/mlx-examples#1093

Update @lenml/tokenizers for tokenizer fixes

Update tokenizers to transformers.js@3

Abort generation on Ctrl-C

Run llama 3.2 in CI

Update README for the new APIs and new models

No more need to limit memory cache

Fix errors in converted code

Expose APIs

Fix issues with tokenizer of llama3.1