llama.cpp

Allow quantize to only copy tensors, other improvements

#2931

Merged