* llama : add token support to llama-grammar * fix inverse token comment * refactor trigger_patterns to replay tokens instead of the entire string * add token documentation * fix test-llama-grammar * improve test cases for tokens