Run unit tests and vLLM-main tests in prebuilt containers
tests.yaml: run in pytorch/pytorch:2.13.0-cuda13.0-cudnn9-runtime and install
into the image's Python, reusing its torch.
vllm_main_tests.yaml: run in the vllm/vllm-openai nightly image instead of
compiling vllm main from source (~1h) and installing the CUDA toolkit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
