Fix non-empty meta when reading from shapefiles

[fbunt]

Closes #302

[fbunt]

I added tests for the other reader types to be safe but I can pull them out if needed.

[fbunt]

Could someone review this?

[martinfleis]

Sorry for delays! This looks good. Can you also add a test that verifies that the erroneous line you had in #302 indeed works now?

[fbunt]

Sure!

[fbunt]

I added a test module for IO integration tests. Let me know if this is overkill.

[martinfleis]

Fine with me, thanks!

[martinfleis]

Actually, the minimal env does not have pyogrio. Can you add skipif to avoid the failure we see now?

[fbunt]

Fixed

[fbunt]

Is it possible to do a patch release to get this out there?

[jorisvandenbossche]

@fbunt will look into doing a small bug-fix release tomorrow!

[fbunt]

Thanks!

[jorisvandenbossche]

FWIW, I was wondering if this should be considered a bug in dask-expr (that their set_index should not pass such meta object to pandas' set_index), but so dask does document _meta as always being empty (https://docs.dask.org/en/stable/dataframe-design.html?highlight=_meta#metadata). So regardless of this fix being a workaround for what dask-expr set_index does, creating empty meta objects seems to be what is expected.

@fbunt and thanks for the fix!

[fbunt]

Yeah it's tricky. I wasn't sure if that's how the rest of the _meta machinery works in dask/dask-expr.