Job splitters should retain original geometries

[VincentVerelst]

@kvantricht, updated both split_job_hex and split_job_s2grid so that they now:

• Retain the CRS of the original geodataframe
• Retain the geometries of the original geodataframe
• Use Web Mercator instead of lat lon to calculate the centroids, so that the warnings are gone

Also added unit tests for both.

[kvantricht]

Just one remark on the speed of reprojecting s2 grid.

[kvantricht]

is this not a slow step for the entire s2 grid?

[GriffinBabe]

@VincentVerelst Maybe if this is a speed issue you could upload a second s2 grid in the artifactory with the CRS reprojected, and then change the load_s2_grid() function in gfmap to accept a parameter web_mercator: bool = False

[kvantricht]

May be worth the effort indeed.

[VincentVerelst]

Thanks for the suggestion, I'll already update that.

Do have to note that the largest bottleneck is typically not the CRS conversion of the s2 grid, but rather the CRS conversion of the dataframe to split itself.

For example, for 2018_BEL_LPIS-Flanders_POLY_110 the old job_splitter takes 10s to run, while the new one takes 20s.