Tests: fixed-tolerance over strict NumPy array comparison

In the test suite, for direct comparisons of arrays where some operation has been applied to either or both arrays that results in floating points and therefore is subject to floating point precision, it would be more robust to check the arrays are equal or otherwise within a given tolerance, over comparing for strict element-wise equality e.g. of some arrays cf_data.array and numpy_array as applied often throughout the suite via some statement ultimately similar to:

assertTrue((cf_data.array == numpy_array).all())

This came to my attention because a specific test test_Data_exp failed on one of the CI jobs, whilst passing locally and on other jobs for a different OSs, where changing to a fixed-tolerance comparison namely to a numpy test method:

numpy.testing.assert_allclose(a.array, b)

meant it passed consistently.

I am aware cf-python has its own tolerance handling for equality testing but here I am referring to cases where we are testing the values of numpy arrays underlying the field constructs or data, such that cf-pytohn methods are unavailable. And even if they were, I think it is preferable to not use that in testing as the behaviour of the tolerance handling would add another dependent variable to the 'experiment' of each test.

Start by locating test_Data_exp and the other test cases that directly compare NumPy arrays with strict element-wise equality. Review the existing fixed-tolerance comparison using numpy.testing.assert_allclose, then update applicable floating-point comparisons and run the affected tests; done means they remain consistent across CI environments.