[rql2sql] closes #1832859: fake HAVING terms w/ EXISTS terms
there are cases where we want to use the HAVING trick on terms that should
belong to an EXISTS subquery. Currently , this causes bad sql. We should fix
that until we've a proper rql grammar avoiding the need for this trick.
Notice at some point, we may want **actuall** HAVING clauses and GROUPBY used
in EXISTS subquery, but that's another story.