llama.cpp

Override SSM_A op for Qwen3 Next to reduce splits

#17587

Merged