Sort nodes above Window/Group By Aggregations

[max-hoffman]

We appear to push Sort nodes belowGroupBy and Window in some cases. In general window functions are responsible for sorting the input buffer once for partitioning, and then a second time for restoring the original sort order.

tmp3> create table mytable (a int primary key, i varchar(20));
tmp3> explain SELECT COUNT(*) c, i AS foo FROM mytable GROUP BY i ORDER BY i DESC;
+---------------------------------------------------+
| plan                                              |
+---------------------------------------------------+
| Project(COUNT(*) as c, foo)                       |
|  └─ GroupBy                                       |
|      ├─ SelectedExprs(COUNT(*), mytable.i as foo) |
|      ├─ Grouping(mytable.i)                       |
|      └─ Sort(mytable.i DESC)                      |
|          └─ Projected table access on [i]         |
|              └─ Exchange(parallelism=12)          |
|                  └─ Table(mytable)                |
+---------------------------------------------------+

We should preserve this behavior in the analyzer when the grouping/partitioning scheme matches the sort order, but otherwise Sort should live above GroupBy.