fix: add ak.array_equal to NEP18 overrides, documentation, and add one more test

[tcawlfield]

• Now overrides numpy.array_equal
• Adding link to toctree
• Unit test checks numpy override

[jpivarski]

This is good; I agree that the heading :caption: should be "approximation and comparison".

The first set of test failures are due to the fact that Awkward only supports native-endian numbers, and you have some tests that make big-endian unsigned int32. I don't know what the second issue is.

Once those are fixed, this is mergeable!

[tcawlfield]

The first set of test failures are due to the fact that Awkward only supports native-endian numbers, and you have some tests that make big-endian unsigned int32. I don't know what the second issue is.

Oh, a regression failure! I'm seeing the first of these errors on my own system once I run all tests. Hmm. The failure is here.

almost_equal is getting the layout of both inputs sooner than it really can in the case that one array is a pure numpy array. Maybe I need to add some earlier tests of that condition -- if one (or both I suppose) array is raw numpy (or cupy or jax etc), then do a simpler set of tests. I could also try/except and ... no, that won't work. Or maybe only run to_layout(...) on Array instances, and only to_packed() on particular layout classes?

I'll work on this.

[jpivarski]

This makes sense (the update in ec93fd2): "same backend" is a broader criterion than someone is likely to be asking about when they ask whether two arrays are equal. Having it raise an error, like all of the other functions, is good for consistency.