card: live demo + ONNX note

ONNX export (fp32 + int8) for Transformers.js / in-browser inference

v1 model card: benchmark table (tfidf / zero-shot / qwen3 / ours), usage, training, limitations

v1 weights: bge-small-en-v1.5 fine-tuned on coa-mapping-synthetic (acc@1 0.883 on COA-Map-Bench v1)

Model card: COA mapper (in training)

initial commit