Data.mask: migrate to Dask

[sadielbartholomew]

Convert the mask managed attribute towards #182.

[davidhassell]

Hi Sadie, I'm afraid that I favour the existing behaviour, which I also see as being consistent with numpy. For a numpy array a, a.mask returns a boolean numpy array; similarly for a cf.Data object d, d.mask returns a boolean cf.Data object. I'm not sure what d.mask should return if not a cf.Data object. The cf.Data object mask is lazy (e.g. hasn't triggered a read from disk at definition time), and also supports extra features (such as cyclic axes) which are not supported elsewhere.

The following also works:

>>> import cf
>>> d = cf.example_field(0).data
>>> a = d.mask.array
>>> (a == d.mask).all()
True

i.e. we don't need the .array bit for everything to work through.

[sadielbartholomew]

Hi @davidhassell, fair enough, thanks for explaining the context - I think in hindsight I was getting confused over the nature of mask and after your explanations I agree about the behaviour it should have. I will update this PR accordingly shortly.

[davidhassell]

Hi Sadie, thanks - I've suggested a different tack on the main method, we can discuss this afternoon, perhaps.

[sadielbartholomew]

All feedback has now been addressed. Triggering the CI jobs via open-close for a final sanity check...

[sadielbartholomew]

All good: local test passes, CI jobs pass, & all feedback has been addressed, including adding a new unit test. Clearance to merge assuming similar conditions from David given externally prior to his taking leave, so merging now.