* Remove unused columns from subqueries

This commit reduces the output columns for existential subqueries.
For existential subqueries, we add a groupby operation on a key of the
outer table, to simulate a semi-join. The grouping columns must be a key
of the outer table, plus any columns that are required by the parent
node.

In an earlier commit, we pruned away unused columns in a query, but
PcrsOutput() could still sometimes contain such unused columns. Our fix
makes sure we only add used ColRefs to the grouping columns, to avoid
problems when we generate and execute a query plan.
Co-Authored-By: NHans Zeller <hzeller@pivotal.io>