ENH: Implement partitioning based on hilbert distance

[tastatham]

Implements partitioning Dask-GeoPandas based on calculated hilbert distances - based on the hilbert_distance function in #70.

Details:

• Calculates the hilbert distance
• Adds column using calculated distances
• Sets index/re-shuffles data using new column

[martinfleis]

Let's talk about this tonight. I would maybe prefer a different API, more generalisable.

Let's assume we have hilbert_distance and geohash functions based on which we can spatially repartition gdf. I would then prefer to have a single method (spatial_shuffle?) that can consume both.

# using hilbert
ddf = ddf.spatial_shuffle('hilbert')

# using geohash
ddf = ddf.spatial_shuffle('geohash')

This API can be extended by other methods without the necessity to have two methods for each algorithm (esp. since the repartition_ methods will be largely the same).

[jorisvandenbossche]

I think we might also need APIs at two different levels: a high level one like the spatial_shuffle above that encapsulates both calculating the information (eg hilbert distance, geohash) and doing the actual shuffle, and a "lower level" repartitioning function based on pre-calculated data. For the hilbert case, this might simply be the existing dask set_index, but for geohash this is basically repartitioning based on a discrete attribute (-> #61)

[martinfleis]

Replaced by #104