PyTorch-native TPU back end for vLLM

This project provides a PyTorch-native TPU backend for vLLM, potentially simplifying LLM inference on TPUs. A practical adoption path is to inspect dependency a