[ExecuTorch][WebGPU] SDPA: branchless aligned/tail loads in the QK/AV kernels#20493

meta-codesync[bot] merged 3 commits into

gh/JulianCloudNTH/63/basepytorch/executorch:gh/JulianCloudNTH/63/basefrom

gh/JulianCloudNTH/63/headpytorch/executorch:gh/JulianCloudNTH/63/headCopy head branch name to clipboard