llama.cpp

ggml: support quant scales for fp8 and nvfp4

#28898

Open