generate_terrain: cupy case, dask numpy case

[thuydotm]

No description provided.

[thuydotm]

@kiliakis I've refactored the code and added some tests. It happened that the results when running with vanilla numpy, dask numpy, and cupy backed DataArrays are not the same. Similar things happened when running with the original code you added. Can you investigate and make sure the results match? I can help to take a look into the dask case.

[kiliakis]

@thuydotm The perlin noise function (that is also used by generate terrain) uses a random number generator (see here for example: https://github.com/makepath/xarray-spatial/pull/555/files?file-filters%5B%5D=.md&file-filters%5B%5D=.py#diff-4599a7f9c05a7872e774fd7d4fa503515a3e5eecbed9d36ed379bb2266491a62R71).
The numpy implementation uses numpy.random and Cupy uses cupy.random. These functions do no return the same sequence of random numbers.
So my guess is this is why we don't see the same results. To see the same results, we will have to modify the code to not shuffle the p array (see here: https://github.com/makepath/xarray-spatial/pull/555/files?file-filters%5B%5D=.md&file-filters%5B%5D=.py#diff-4599a7f9c05a7872e774fd7d4fa503515a3e5eecbed9d36ed379bb2266491a62R76). But this is something that can be used for testing only.

[thuydotm]

@kiliakis I merged your PR and made code changes to ensure the same results when running with numpy, dask and cupy. Going to merge this PR to main branch. Please feel free to open new PRs if you see anything to improve.