[AMD] Optimize Qwen3-VL decode - fuse QK-norm + 3D mRoPE + KV cache write#21458

HaiShaw merged 6 commits into

sgl-project:mainsgl-project/sglang:mainfrom

yctseng0211:fused_qk_norm_mrope_decodeyctseng0211/sglang:fused_qk_norm_mrope_decodeCopy head branch name to clipboard