Added a pure numba hillshade that is 10x faster compared to numpy

[alex-soklev]

The new code uses less context switching and data transfers, minimizes communication overhead and removes the cupy dependency. I've made it run by default, if cuda is available. Otherwise if falls back to numpy

This change addresses issue #541

[brendancol]

• change to I/O to cupy arrays

[brendancol]

@thuydotm can you review this and get back with any comments?

[thuydotm]

Thanks @a7az0th!
I think we can now remove the old run_cupy function and replace it with the run_numba since the old code is no longer called. run_numba should be renamed to run_cupy to be consistent with our coding convention. Otherwise looks good to me.

[alex-soklev]

@thuydotm I've updated the source according to your feedback. I believe we can merge this.

[thuydotm]

Merged, thanks!

[alex-soklev]

Thank you, too!