[Cherry-Pick][Optimization] Use triton qk_norm both in Prefill and Decode.(#7213)#7305

EmmonsCurse wants to merge 1 commit intoPaddlePaddle:release/2.5PaddlePaddle/FastDeploy:release/2.5from

EmmonsCurse:cherry-pick/7213/release/2.5EmmonsCurse/FastDeploy:cherry-pick/7213/release/2.5Copy head branch name to clipboard