Write specific functions to collapse/rebuild input/outputs as Raster, np.ma.masked_array or np.ndarray consistently

[rhugonnet]

We have tons of "wrapper" functions that can take either Raster, np.ma.masked_array or np.ndarray as input, but then only work on the np.ndarray (and possibly some metadata), and for output rebuild a Raster, np.ma.masked_array or np.ndarray to match the input.
Right now this is done semi-manually everywhere, we need to have consistent input/output functions for those conversions. Might be triggering issues in #291 due to GlacioHack/geoutils#265.

[erikmannerfelt]

That's a good idea. A small problem right now (or an unintended feature) is that all metadata are lost after raster recreation. Only the CRS and transform are identical to the input.

[adehecq]

@rhugonnet, could you add here the few lines of code that are repeated each time? Is it more than just 1-2 lines of code?

[rhugonnet]

Yes, the 10 lines in Coreg.apply() for "reconstructing" for example: https://github.com/GlacioHack/xdem/blob/main/xdem/coreg/base.py#L1160

[rhugonnet]

We could do the same to construct masks from a Vector, a Mask, an array, etc... It is used everywhere!
Something like this: https://github.com/GlacioHack/xdem/blob/main/xdem/spatialstats.py#L593