llama.cpp

fix CUDA CI by using a windows-2019 image

#7861

Merged