Pinned copy of meta-llama/Llama-3.2-1B tokenizer.json, for huggingface/tokenizers tk-convert oracle test (byte-level BPE model, gated on the Hub so mirrored here ungated).
Add llama-2-7b-chat-hf tokenizer.json (canonical, array-safe merges) for pipeline oracle (#6)
