llama.cpp

ggml : check cuda and metal argsort limits and add test

#16323

Merged