Add more aggregate test cases for ColumnarIndexScan

Cover query shapes that were missing from the test suite:
- DISTINCT on a segmentby column (Agg without Aggrefs)
- count(*) without GROUP BY
- aggregate only in HAVING (not in SELECT)
- GROUP BY ordinal position
- GROUP BY () empty grouping
- aggregate wrapped in COALESCE
- aggregate above a JOIN
- ordered-set aggregate (percentile_cont)
- bool_and / bool_or
- partition-wise aggregate
- Aggref inside a SubLink in the outer tlist (inner aggregate belongs
  to a separate Agg and must not be touched by the outer rewrite)
- HAVING with non-pushable expression on grouping columns (qual stays
  with the Agg instead of being pushed to the scan)
- same Aggref node referenced from multiple tlist positions
- GROUP BY column not in SELECT (appears only as resjunk in Agg tlist)
- WHERE mixing segmentby and non-segmentby columns
- HAVING NOT (Aggref ...) and HAVING Aggref IS NULL (mutator must
  recurse through BoolExpr and NullTest)
- GROUP BY with explicit collation (grouping expression is a
  CollateExpr, not a bare Var)
- self-join on the same hypertable (Agg sits over a Join, not over
  a ColumnarScan)

Author: svenklemm

tsl/src/nodes/columnar_index_scan/columnar_index_scan.c

@@ -739,11 +739,7 @@ insert_columnar_index_scan(Plan *plan, void *context)
 	}
 
 	/*
-	 * GROUPING SETS / ROLLUP / CUBE produce an Agg with a non-empty chain of
-	 * additional Aggs (and Sorts) that share the same input. The rewrite below
-	 * remaps only this Agg's targetlist and grpColIdx, so chained Aggs would
-	 * end up referencing the pre-rewrite column layout. Refuse the rewrite in
-	 * that case.
+	 * GROUPING SETS / ROLLUP / CUBE currently not supported.
 	 */
 	if (agg->groupingSets != NIL || agg->chain != NIL)
 	{