switch back testing of irfftn shapes

[ev-br]

I believe this is a correct interpretation of https://numpy.org/doc/stable/reference/generated/numpy.fft.irfftn.html

cross-ref gh-301

[asmeurer]

Good to always check the standard itself (because sometimes there are typos in the standard, and sometimes it intentionally deviates from numpy) https://data-apis.org/array-api/latest/extensions/generated/array_api.fft.irfftn.html#irfftn. In this case, I think it also matches.

[asmeurer]

But maybe you could double check yourself if it looks like the standard agrees with this. If so, let's merge this. If not, most likely we need to issue a fix to the standard (for fft it was mostly intended to match numpy; I already made a lot of fixes at data-apis/array-api#720, but I wouldn't be surprised if I missed something)