Found this while investigating a test failure on #26628, the following two queries have different results even after manual reordering of result rows.

> SELECT (i.keys).n 
      FROM (SELECT information_schema._pg_expandarray(indkey) AS keys FROM pg_index) AS i ;

> SELECT (i.keys).n 
      FROM (SELECT information_schema._pg_expandarray(indkey) AS keys FROM pg_index) AS i 
      ORDER BY 1;

The specific symptom is that if all the values of n are different in the 1st query, then some of them will become duplicated in the 2nd query.

I suspect some sort node badness i.e. mutating the tuples provided by project set in-place.

@solongordon you may have seen this without realizing