ENH: add Parquet/Feather IO

[jorisvandenbossche]

Adding read_parquet/read_feather and to_parquet/to_feather functions, so we can work from large data from disk (and not just from existing geopandas.GeoDataFrames).

An initial version could be a relatively simple wrapper of the dask + geopandas functionalities, I think.

(and later we should look into spatial partitioned datasets, but that first requires spatial partitioned dataframes)

[jsignell]

This sounds good. Once we have spatially partitioned datasets we can determine a metadata spec that the reader can look for in the file.