Add SGLang Qwen3-8B baseline benchmark harness #91

Job	Run time
UnitTest (cuda, self-hosted, CUDA, nvcr.io/nvidia/pytorch:26.03-py3, --privileged --ipc=host --gp...	14m 28s
UnitTest (rocm, self-hosted, ROCM, rocm/pytorch:rocm6.2.3_ubuntu22.04_py3.10_pytorch_release_2.3....	1h 11m 3s
	1h 25m 31s

Provide feedback