Remove explicit use of Masked Array types

[Cadair]

Provide a general description of the issue or problem.

It's unclear to me why we explicitly use np.ma.masked_array and the dask equivalent. The problem with doing this is we are hard-coding the masked array handling in rebin to be numpy and dask only, excluding other things such as cupy.

We could just carry around two arrays, one for the data and one for the array? @DanRyanIrish