🐛 Describe the bug

In #89332 I added a new sample input, and as a result, I had to disable cat testing on MPS entirely, because only that sample input was computed incorrectly on MPS. It would be nice if I could have only xfail'ed that sample input.

In general, running all sample inputs in a single test is problematic, as you cannot easily tell if there is only one sample input that is problematic / which sample input is problematic. Creating a separate test per sample input feels like it would be better.

Versions

master