[dtensor] disable gpu tests in op db first (#92611)
There seems to be some issue with the cuda tests where our CI aren't
capturing those failures (probably because of lacking 4 GPUs in CI
environment). Disabling it first and debug later
see https://github.com/pytorch/pytorch/issues/92343
Pull Request resolved: https://github.com/pytorch/pytorch/pull/92611
Approved by: https://github.com/XilunWu
