llama.cpp

kv-cache : refactor + add llama_memory_state_i

#13746

Merged