Fixed iris.util.array_equal to avoid issues with fully masked arrays

[esc24]

The following fails on upstream/master:

>>> import iris
>>> cube = iris.load_cube(iris.sample_data_path('space_weather.nc'), 'electron density')
>>> res = cube + cube

producing
ValueError: This operation cannot be performed as there are differing coordinates (longitude) remaining which cannot be ignored.
I tracked this down to the fact the longitude coordinate is fully masked (I know masked coords are not supported in Iris, but that can wait for another day). The problem stems from:

>>> lon = cube.coord('longitude')
>>> lon == lon
masked

This PR fixes the problem by altering iris.util.array_equal so it ignores any mask if present. This matches the behaviour of numpy.array_equal as stated in the existing iris.util.array_equal docstring. You could argue that we should modify the function to handle masked_arrays properly (that is open to various interpretations), but the chosen approach fixes the problem, is simpler, and reflects numpy behaviour.

Note that the change also means that iris.util.array_equal now matches numpy.array_equal in other ways too such as handling other types via np.asarray().

[pelson]

Do we not need to give this a specific name? (Update: No, it appears not).

[pelson]

Given the reason this function exists at all, I'd be more comfortable if there was a test checking the string equality (with strings of different lengths).

[pelson]

Other than that, the implementation certainly is much neater. 👍

[esc24]

I've added some more tests @pelson.