FIX/MNT: Simplify project geometry handling

[greglucas]

The MultiLineString return type was a plain list, but it should be an empty MultiLineString to be consistent with the other return types. Additionally, all geometry constructors take empty lists, so just use that rather than special-casing the returns.

Closes #2406

[greglucas]

Interesting, the geom_type has changed from GeometryCollection to MultiLineString. So previously, this would have failed before even making it into the transformation because we don't know how to project GeometryCollections. Should we ignore this test with shapely<2, or other suggestions for how to handle it?

[dopplershift]

I think it's important to know whether any of those classes accepted empty lists on Shapely < 2? It's not clear to me whether the test is checking code paths that we could even hit before in practice.

Essentially, after this change: do we actually still support Shapely < 2?

[greglucas]

Yes, we still have a minimum test that pulls in Shapely 1.8.

https://github.com/SciTools/cartopy/actions/runs/9726193384/job/26844355302#step:4:1

Shapely 2.0 changes the geom_type return to be more correct. Not a regression for us, just additional things that will get projected properly now that we get for free. It just impacts the test where this fails without shapely 2.0 and was also failing before this update.

[dopplershift]

This looks good to me then.