Add tokenizer.json (transformers.js fast tokenizer)

Initial upload: INT4 ONNX (encoder + prefill + step) + tokenizer + examples + inference.py

initial commit