Generate expected data for integration tests as feather files

[Jefffrey]

Relates to #66

Use PyArrow to read ORC files and write the data as Arrow feather files. This is to have more robust equality checks instead of relying on JSON (which needs to be parsed back to Arrow first).

Generating the expected files is a once off activity, relevant script included.

[progval]

Would it make sense to make build.rs run the Python script, so .feather files don't have to be committed to Git?

[Jefffrey]

Would it make sense to make build.rs run the Python script, so .feather files don't have to be committed to Git?

Hmm that's a good point, I didn't consider that.

One caveat is we'd need to run in a Python venv or use a Docker container to handle the pyarrow package requirement in a robust manner

[Jefffrey]

Created an issue for the above

#74