Add optional MLX backend for Apple Silicon (backend='mlx' or 'auto'; ONNX stays the default)

Rename router.py to laya_intent_router.py, Router to LayaIntentRouter

router.py: CLI falls back to Hub weights when run outside the repo

Single-file ONNX (fixes from_pretrained), better model card

Laya Intent Router 150M: int8 ONNX, shortlist embedder, router.py, model card

initial commit