fix: register Numba in test

[agoose77]

This PR recognises that Numba isn't being registered for a specific test, and fixes it!

[agoose77]

@jpivarski I simplified this to only address a missing ak.numba.register_and_check().

[codecov]

Codecov Report

Merging #2879 (d36a8b0) into main (9400780) will not change coverage.
The diff coverage is n/a.

Additional details and impacted files

[ianna]

@agoose77 - I think we also need to check on the numba version greater or equal to 0.58 to avoid specifying the types for vectorise. Perhaps, adding a warning in case the version does not support it yet?

[jpivarski]

Looks good.

About the minimum Numba version: it's true that @nb.vectorize without explicit arguments won't work before a certain Numba version, but that could be described as something missing from Numba, rather than something missing from Awkward.

In addition, @nb.vectorize is not a very popular function. This plot comes from static analysis of all the non-fork repos that import numba (or from numba import):

(If Numba is imported as another name, like nb, this includes nb.vectorize, and the nb.jit and nb.njit have been folded such that nb.jit with nopython=True counts as nb.njit only.)

So it doesn't seem necessary to me for Awkward to require a much newer Numba version just for vectorize. Since the Numba dependency is not managed by pip/conda (it's an optional dependency), dealing with mismatched version numbers is more annoying, as it's something the user must do manually.

[jpivarski]

All of the tests pass. @ianna, if you agree, you can do the merge. If not, let me know your reasons and we'll talk!

[ianna]

Thanks, @jpivarski! Yes, I agree, let's merge it!