AMD/ROCm implementation and testing (llama.cpp/vLLM)

[ericcurtin]

On completion, a person will have tested and implemented ROCm on llama.cpp and vLLM successfully with ROCm acceleration. Our ROCm vLLM container should install using something similar to this:

https://www.phoronix.com/news/AMD-ROCm-vLLM-Wheel

Requires a well supported AMD/ROCm card to do this work.

[doringeman]

This is a PR that can be tested: #638.