Resolve MultiPolygon with holes

[mattijn]

This PR fix #148 and #167.

I've changed the _resolve_bookkeeping() function. This function replace the geom ids with the corresponding arc ids. Before the corresponding arc ids were collected without considering the shape of the geom id. Now we take the arcs_in_geom as base and set the arc_ids for the corresponding arc in the same object.

This gives the right nesting for MultiPolygons and MultiPoint, but for other feature types the arcs become nested one level too deep. By updating the _resolve_arcs() function to unnest nested arcs for the other features, I could make all the tests pass. Except one (test_hashmap_geomcol_multipolygon_polygon).
Before the approach was done reversly. All single features were fine, but the multi features were nested one level deeper.

Upon inspection of the failed test, I realized that this covers the issue as is described in both linked issues, but was asserted wrongly..

Now for this test the input MultiPolygon is also a MultiPolygon in the output.