backward_compatibility_check_test doesn't play well with reverts

[ezyang]

bc test is broken on master because of this revert commit: #29329

In general, it's ok to break BC if we're reverting a diff. It would be nice if the test knew something about this.

cc @ezyang @houseroad

[zou3519]

This pattern happened again today:

The pytorch_linux_backward_compatibility_check is currently broken on master. This is because:

1. we merged a PR to add logical_and
2. The logical_add commit broke master, so we reverted it
3. The backward compatibility check noticed that logical_and didn't exist anymore so it started complaining.

(thanks to @albanD for noticing the breakage and pointing it out)

I think these "add a new operator" and then "revert that new operator" workflows are probably going to be relatively common so it would be nice for the BC test to understand these.

[houseroad]

If we add a timestamp to each function schema, it's possible to have a buffer window. For now, I would suggest to add these reverted ops to the whitelist to make the CI happy.

I have created #30537 to fix current breakage in op BC check.

[huydhn]

This is a known issue but I think it's minor. My attempt to fix this #92708 was dropped because of the lack of interest. So I think we can close this issue.