use chunksizes and check for empty geometries and empty dataframes

[alpha-beta-soup]

Closes #17 by changing the interface; it's more sensible to give a chunksize (number of rows per partition after cutting and spatial sorting) than it is to target a number of partitions.