llama.cpp

fix(graph): remove duplicate wo_s scale after build_attn (Qwen3, LLaMA)

#22421

Merged