Port natural neighbor interpolation to Shapely

[dopplershift]

Natural Neighbor interpolation uses a lot of computational geometry operations, all currently implemented in pure Python. Shapely provides some better primitives and operations that might allow us to eliminate some code, and benefit from the fact that much of it is compiled underneath.

They also have (in >=1.8) implementations of Delaunay triangulation/Voronoi polygons that we might be able to use instead of scipy's--if that gains us something.

[dopplershift]

This would had a dependency on Shapely to our list, though that's already largely in our list due to Cartopy (though it's optional), as well as declarative using it (in checking types).

[dopplershift]

When doing this, it might be useful to see how our implementation compares with this: https://gwlucastrig.github.io/TinfourDocs/NaturalNeighborTinfourAlgorithm/index.html#is-it-accurate