[GEN-2381] Pandas handling of nullable cells

[BryanFauble]

• Verified that the system correctly handles null values during upserts and that the data is stored and retrieved accurately.

Problem:

• When querying for, inserting, or updating cells of data that can be nullable things start to break as described in https://sagebionetworks.jira.com/browse/GEN-2381
• Using convert_dtypes introduces int64 data not serialized by json error and attributes not matching error in integration tests such as StringDtype vs object.

Solution:

• Use the suggestion from @danlu1 to use both the convert_dtypes and the dtype argument when reading in a CSV to pandas DF

Testing:

• Unit tests and integration tests are extended and run though successfully.
• More testing within Genie is needed

[thomasyu888]

For discussion: Should expected dataframes in these tests be created with the dtypes set so that we are testing the fact that convert_dtypes is being done?

The concern here is if there are issues with "convert_dtypes" function, it won't be caught because we are using the function itself.

[danlu1]

The only reason I added convert_dtypes is to make data types match between columns. But I forgot that we can use check_dtype=False instead. I would prefer to use check_dtype because it would be a quicker fix.

[rxu17]

It's still expected that the result of calling query will return object type for something that is integer type and has nulls.

Then using convert_dtypes() would convert it to the correct pandas dtype (in this case Int64)What we are trying to prevent was that previously query would return something from a table that is integer with nulls -> float.

If that is the case, then i think this is good to go for the genie scenario.

[rxu17]

LGTM! Just a comment