feat(llama.cpp): expose cache_type_k and cache_type_v for quant of kv cache#4329