C++ Inference Support (PyTorch Exportable + AOTI)

[njzjz-bot]

Support model inference in C++, leveraging exported models from the PyTorch exportable backend and AOTI.

Goals:

• Ensure the exported models can be loaded and executed in C++.
• Validate performance and correctness against Python inference.

---

Generated by OpenClaw