Fix ndistinct estimates with system attributes

When estimating the number of groups using extended statistics, the code
was discarding information about system attributes. This led to strange
situation that

    SELECT 1 FROM t GROUP BY ctid;

could have produced higher estimate (equal to pg_class.reltuples) than

    SELECT 1 FROM t GROUP BY a, b, ctid;

with extended statistics on (a,b). Fixed by retaining information about
the system attribute.

Backpatch all the way to 10, where extended statistics were introduced.

Author: Tomas Vondra
Backpatch-through: 10

src/backend/utils/adt/selfuncs.c

@@ -3989,10 +3989,10 @@ estimate_multivariate_ndistinct(PlannerInfo *root, RelOptInfo *rel,
 
 			attnum = ((Var *) varinfo->var)->varattno;
 
-			if (!AttrNumberIsForUserDefinedAttr(attnum))
+			if (AttrNumberIsForUserDefinedAttr(attnum) &&
+				bms_is_member(attnum, matched))
 				continue;
 
-			if (!bms_is_member(attnum, matched))
 			newlist = lappend(newlist, varinfo);
 		}
 

src/test/regress/expected/stats_ext.out

@@ -260,7 +260,7 @@ SELECT s.stxkind, d.stxdndistinct
 SELECT * FROM check_estimated_rows('SELECT COUNT(*) FROM ndistinct GROUP BY ctid, a, b');
  estimated | actual 
 -----------+--------
-        11 |   1000
+      1000 |   1000
 (1 row)
 
 -- Hash Aggregate, thanks to estimates improved by the statistic