llama.cpp

fix: crash when calling `llama_state_get_size` on a context without a KV cache

#13542

Merged