BUG: fix read_parquet with dask>=2022.12.0

[the-matt-morris]

closes #225

Updates method signature of the following so that they are compatible with all versions of dask, including dask>=2022.12.0:

• dask_geopandas.io.arrow.ArrowDatasetEngine._arrow_table_to_pandas
• dask_geopandas.io.arrow.GeoDatasetEngine._arrow_table_to_pandas
• dask_geopandas.io.parquet.GeoArrowEngine._create_dd_meta

It's been a little while since I played around with this, but I had written previously that these lines in dask are problematic, as it calls _arrow_table_to_pandas, where it wasn't called prior to dask==2022.12.0.

At the time, I thought we may need an update to GeoDatasetEngine._arrow_table_to_pandas to handle this, but I feel like someone with a better grasp on this library's internals may have a better idea. Either way, it is known that this PR in its current state is an attempt to solve the problem, but it doesn't succeed entirely in doing that, as you'll see with one of the unit tests failing.

[jorisvandenbossche]

Either way, it is known that this PR in its current state is an attempt to solve the problem, but it doesn't succeed entirely in doing that, as you'll see with one of the unit tests failing.

There are two failing tests at the moment. One will be fixed by #229 (and is unrelated to the parquet IO), and the other is a case that should error but is giving a wrong error message. But that indeed seems to be related to the lines you link to (_arrow_table_to_pandas is being called with an empty table, and that is not working with our implementation). I will take a look at that.

[jorisvandenbossche]

And thanks for opening the PR!

[the-matt-morris]

You're right, I was confused as to why that other test was failing, glad there's a PR for that one already. Thanks for checking it out!

[jorisvandenbossche]

We can probably further improve this and make the integration more robust, but let's already merge this so we can have green CI and have this running with latest dask.

Thanks again @the-matt-morris!