llama.cpp

vulkan: change graph_compute to be async and enable get_tensor_async

#17158

Merged