TurboQuant AMD Linux ports start surfacing

This looks less like a polished release and more like an active porting race around a fresh research result. AMD users can experiment now, but the path runs through forks, custom builds, and GPU-specific caveats.

• –A clean llama.cpp HIP/ROCm port exists in `domvox/llama.cpp-turboquant-hip`, with build instructions and explicit `gfx1100` support for the RX 7900 XTX.
• –Another fork, `spiritbuun/buun-llama-cpp`, advertises AMD support via HIP, but it is still a broader CUDA-first codebase rather than a turnkey AMD release.
• –The strongest AMD evidence so far is still community benchmarking and discussion, not a stable upstream merge.
• –For 7900 XTX owners, the practical issue is less TurboQuant itself and more ROCm/HIP build friction plus backend-specific bugs.
• –The signal here is real developer interest: people are already testing long-context KV-cache compression on AMD, which suggests the ecosystem is moving fast even if the tooling is rough.