Feature Request / Improvement

When writing data using Python, we want to leverage external dependencies that write the actual data. This is because there are many libraries out there that do a great job, and we don't want to reinvent the wheel.

I think we can mirror the read path, where we use pq.read_table, we can use pq.write_table, and leverage the metadata_collector to collect the statistics that need to be stored in the ManifestEntry.

To make the PR small, I would be to start with unpartitioned tables only, to avoid having to construct the right path.

Query engine

None