Add tests+functionality to assess the interchangeability of Rasters and arrays

[erikmannerfelt]

Brought up by @rhugonnet in #175.

Here is the conversation:

Actually, a big question raised by this:
With GlacioHack/GeoUtils#210 and this PR, we now have many use cases where we can use np.ndarray and Raster objects interchangeably. But there is still several functions for which it does not work, right? Such as terrain attributes.

Should we aim to have xdem working with np.ndarrays and Raster objects interchangeably? If yes, what is left to do? (I think that, @erikmannerfelt, you have the best viewpoint on all this!)
Also, it should be crystal clear when a Raster can be provided simply as a convenience and a np.ndarray would suffice, or when georeferenced data is absolutely necessary. It will get blurry for the users if we allow to provide both everywhere for convenience, so I'm not sure how to tackle this!

Since, geoutils==0.0.3, Rasters implement the array interface, so they should be possible to use interchangeably with ndarrays. The tests don't cover that "should" part, however, and Rasters are not consistently returned if provided as an argument. For now, all functions that show a Raster/RasterType argument/return type hint returns a RasterType if given a RasterType, and those who don't will return an ndarray/masked_array.

Long story short, the only way to validate how interchangeable they are is to include more tests!!