Implemented conversion to a "mapping"

Shapely has a mapping function (https://shapely.readthedocs.io/en/latest/manual.html#shapely.geometry.mapping), which can convert a geometry into a dict (in geojson form):

In [59]: from shapely.geometry import mapping, LineString

In [60]: mapping(LineString([(1, 1), (3, 4)])) 
Out[60]: {'type': 'LineString', 'coordinates': ((1.0, 1.0), (3.0, 4.0))}

In geopandas, this is for example used in the iterfeatures / __geo_interface__ and so also in to_json (a method that can convert to a geojson like string, which doesn’t use fiona / GDAL like to_file, this is already faster as to_file, and can eg be used for in memory serialization to geojson).

From profiling iterfeatures in geopandas (for the countries dataset, so for polygons/multipolygons), the mapping of the geometries takes around 75% of the total time converting the dataframe to a dict.

So this mapping could be a target for (vectorized) optimization. Since it involves creating python dicts and lists, it might be better suited to do this in cython, and also given that, I am not fully sure it will necessarily give a big speedup (experiments will need to tell).

This is related to https://github.com/pygeos/pygeos/issues/65 in the sense that the same dict needs to be created for both.

[aprilmay]

A fast/direct way to generate a geojson representation would be great for using with many visualization libraries (Mapbox, Leaflet, etc). No idea if that’s useful, but gdal has a function: https://gdal.org/api/vector_c_api.html#_CPPv418OGR_G_ExportToJson12OGRGeometryH

[caspervdw]

Closing; this is solved by #413