Fix names and nullability when creating RecordBatch from MapArray

[balbok0]

Description

When creating a RecordBatch with one of the columns being a MapArray, there are issues with naming of sub-columns, as well as nullability of the "elements/key_value" sub-column.

Related Issue(s)

• closes Not matching column names when creating a RecordBatch from MapArray #1257

Documentation

See https://www.rustexplorer.com/b/o7bfm4 for a breaking example.
I initially thought that the issue is specifically for array with 0-length maps, but nullable mismatch happens even when all of the map elements have values.

[rtyler]

Thanks for the pull request! @balbok0 would you be able to take your failing example and turn that into a unit or integration test so we don't have this regression in the future?

[wjones127]

I’d like for us to fix this, but I also hope we can find a way to simultaneously support older PyArrow versions in the Python bindings.

[balbok0]

@wjones127 are there some specific pyarrow versions you have in mind? I wasn't able to find a specific arrow-cpp/pyarrow version that changed the naming scheme.

[balbok0]

Added failing example as a test.

[wjones127]

This generally looks good. I was worried about PyArrow and other languages because the Arrow standard specifies field names as "key" not "keys" and "value" not "values" (arrow-rs deviates from this). But the PyArrow C Data Interface import actually ignores these field names on map types anyways, so it shouldn't matter that they are different. So we can ignore that worry.

I have a few small nits on the tests, but otherwise looks good to go.

[wjones127]

Oh and could you rebase? We just fixed some of the deprecation warnings.

[balbok0]

Thanks for comments! I applied changes and rebased. Lmk if there is anything else that needs to done before merge.