[python] Support return_arrow for various queries

[johnkerl]

Note that ingest-from-arrow is a separate issue and is not covered on this PR.

[Shelnutt2]

What is the purpose of this function? This makes a full copy of fields and then creates a new pyarrow table from python arrays. Why do we need to perform a copy?

[johnkerl]

@Shelnutt2 it's a separate function so as to be called from the ThreadPoolExecutor -- I'm open to feedback if you have a better way.

Or if you're asking why copy at all -- please notice that it does a copy+cast only when the need to do ASCII-to-Unicode. (It's careful to return None if nothing needs conversion.) And this conversion is crucial -- not optional -- since without this user-level queries like 'cell_type == "myeloid cell"' would return nothing since columns would have values like b"myeloid cell" which is != to "myeloid cell".

Please see

• Queryability of dataframe attribute columns #99
• Store to-be-queried obs/var columns as ASCII (workaround) #101
• Extend ASCII-queryability workaround #138
• Correct ASCII-to-Unicode readback for attribute_filter #141

We had extensive discussions on this a couple months ago. We must (can't not) do Unicode-to-ASCII converstions on writes and ASCII-to-Unicode conversions on reads until such time as we have core mods in place which:

• Permit string attrs to be Unicode and be queryable using query conditions
• Permit string dims to be Unicode at all

cc @ihnorton

[Shelnutt2]

This will be followed up separately.