Add column selection

[dacort]

By default, we do a SELECT * to query Athena.

DuckDB can send the projected columns to the extension, which is what the postgres scanner does. We should do similar.

[dacort]

Unfortunately, DuckDB just sends the ids of the projected columns. So we'll have to either maintain the list of columns in a local cache (since we look them up via the Glue Data Catalog) or query the GDC to get the list of columns again.