llama.cpp

llama: Wire up Qwen3.5/Qwen3.5MoE tensors for NVFP4 support

#20506

Merged