PyTorch reverts XPU ops pin update, rolling back C++20 fixes

PyTorch's automated merge bot reverted a commit (e0ac3f51) on May 25 that had updated the torch-xpu-ops pin to pick up C++20 fixes. The revert was requested by developer guangyey, who noted that the update unexpectedly rolled back the XPU common pin to an older version, causing a regression. The original PR #184649 was intended to improve compatibility with C++20 for XPU operations but introduced a dependency error that required immediate reversal.

This incident underscores the delicate nature of dependency pinning in large-scale open-source frameworks like PyTorch. The XPU common pin controls critical low-level operations for Intel's XPU hardware acceleration. The mistaken rollback could disrupt developers relying on the latest XPU features, though the interim solution is to revert to the previous stable pin while the team investigates the root cause. Users tracking PyTorch's trunk should expect a corrected update once the dependency chain is verified.