convert : support mixed-precision ModelOpt models with per-tensor NVFP4/FP8 quantization#20539

CISC merged 3 commits intoggml-org:masterggml-org/llama.cpp:masterfrom

richarddd:fix/nvfp4-mixed-precision-convertricharddd/llama.cpp:fix/nvfp4-mixed-precision-convertCopy head branch name to clipboard