Don't remove mask on no-mask masked arrays

[djkirkham]

Replaces #1575

Removes all instances of checking if a MaskedArray has no mask and replacing it with an ndarray.

[pelson]

Surprised this behaviour isn't tested...

[pelson]

Relates to #2046.

[DPeterK]

@djkirkham these changes definitely need some testing associated with them...

Thanks for taking this on though!

[djkirkham]

I've reinstated this check since the cube is constructed in this method, and the guard only causes it to be unmasked if the input data was not a masked array, which seems sensible.

[djkirkham]

This array is constructed as a masked array at line 223, so if we don't convert it to an ndarray it will output a masked array even if the input is an ndarray.

[djkirkham]

I've reinstated some of the checks I removed. See my comments on commit f63d950

[pelson]

LGTM. I'll cross-check the behaviour and ensure there is an appropriate what's new entry.

[pelson]

Just to confirm that as a result of this PR, the behaviour is unchanged between iris 1.13 and 2.0.

Code to confirm:

from iris.cube import Cube

import numpy as np
import numpy.ma as ma


data = ma.masked_array(np.arange(12).reshape(3, 4), False)
cube = Cube(data)
print(type(cube.data))
print(cube.data)