Try harder to detect obfuscated min/max patterns: the initial pattern was added with D9352 / rL236202. There was a bug fix for PR27137 at rL264996, but I think we can do better by folding the corresponding smax pattern and commuted variants.
For the PR27137 test case, we could further transform that to a mask with the smeared sign bit (ashr/and/not), but I'm not sure if we want to do that in IR?
Is there really no better way to write a testcase for this? I guess there are a limited number of transforms which use min/max formation, but this seems awkward.