llama.cpp

sync : ggml (new ops, tests, backend, etc.)

#4359

Merged