Llama.cpp Now Part of the Nvidia RTX AI Toolkit

I don't want nvidia to pull in projects.

I don't want ROCM to be the only way to use HIP.

Just upstream stuff, I don't want some special proprietary package.