ONNX Runtime as a Portability Layer: Export, Verify, Deploy

Exporting a PyTorch model to ONNX, proving the numerics survived, and running the same graph across execution providers.