MultiPolygon "coordinates" member should support empty Polygon arrays inside of it.

[airbreather]

This was first brought to my attention in #140, which identifies a specific subset ([[]]) supposedly produced by one specific tool (gdal-bin ogr2ogr) and resolves that to do the right thing.

While that PR is a good first step, we should fix the underlying issue more completely and generically, such that an empty Polygon array anywhere within the MultiPolygon's "coordinates" member will be handled properly. Examples for test cases:

1. [[]] is a MultiPolygon that contains one empty Polygon in it (fixed by Deserialize MultiPolygon Array of Empty Coordinates Array #140)
2. [[], []] is a MultiPolygon that contains nothing but two empty Polygons in it
3. [[[[0, 0], [1, 0], [0, 1], [0, 0]]], []] is a MultiPolygon that contains one empty and one non-empty Polygon
4. [[], [[[0, 0], [1, 0], [0, 1], [0, 0]]]] is equivalent to the previous row but in a different order (which matters for the spaghetti that implements this parsing logic)
5. [[], [[[0, 0], [1, 0], [0, 1], [0, 0]]], []] is a combination of both of the previous rows