Add README documenting export, quality numbers, and ORT runtime requirements.

init: Octen-Embedding-0.6B FP16 ONNX (streaming converter, cos=1.000 vs PyTorch ref)

initial commit