Evaluate wider SQL support by fetching data to one node

For SQL queries that Citus currently doesn't parallelize (for example, repartitioned subselects with large table joins), we could pull all related data to a single node and run the query on that node.

This way, if Citus can't parallelize a query, we'd still run the query instead of erroring out on it. The drawback is that these queries would go much slower than ones that get parallelized.

No files, tests, or entry points are named in the issue. First clarify how unsupported SQL queries should be detected and how related data would be fetched to one node; done would mean such queries execute instead of erroring, with the slower behavior understood and tested.

Written by the indexing model from the issue text.