When doing a filesort for removing duplicates (SELECT DISTINCT),
we may need to run another sort afterwards to satisfy ORDER BY.
However, in the instance where we had pushed said ORDER BY down
into the first table, as opposed to the JOIN as a whole, we would
not have a Filesort object for it (it would say nullptr), and forget
to do the final sort.

Change-Id: I5e7e6aa1ffce994efaa98188867cd820c44a4796