llama.cpp

cuda : fix nkvo, offload and cuda graph node properties matching

#19165

Merged