Fix memory explosion

[jnnr]

Reviewed and refined changes from #16

• simplified variable names
• removed unnecessary steps
• used Gregor's own function for zonal sum: aggregate_raster_to_polygon

TODOs

• Maybe parallelize aggregate_raster_to_polygon with dask, too?
• Use the example data (NL) for another test.

[irm-codebase]

@jnnr just ran this on the powerplants module for MNE, MEX and European nations.
Works like a charm.

You can see the moment dask takes over for the European case. Both swap and RAM get close to the max, but never cause a catastrophic failure.

I did not run numerical tests on the produced raster, but an eyeball tests suggests it's good (the data range is the same as in #16 (comment))