cc @ev-br

This update picks up several improvements in the test suite for the 2024.12 array api revision: the updated suite has a full functional coverage of the 2024.12 spec. Thus this PR LGTM.
The failing test is a known issue where numpy deviates from the Array API spec, and so do basically all other libraries:
data-apis/array-api-tests#317
Thus it seems reasonable to just add an xfail similar to https://github.com/data-apis/array-api-compat/blob/main/numpy-xfails.txt#L3 and move on.

One potential follow-up improvement could be to increase the number of hypothesis examples the CI job runs. It currently uses --max-examples=100 which is rather small. This is a kind of important because hypothesis is essentially fuzzing, and a small number of examples means that it does not even probe "interesting" cases.

In array-api-compat, we recently bumped the number of examples to 1000 while keeping the runtime under 5 minutes in data-apis/array-api-compat#321

I'm a bit out of my comfort zone here, but I think I xfailed the failing tests now, and I increased the --max-examples to a modest 500.

LGTM.

Thanks Joren.