MDEV-36321 keys generated on derived tables produce wrong out_rows estimates

record per key statistics on keys generated for derived tables are missing,
causing a fallback to the usage of MATCHING_ROWS_IN_OTHER_TABLE in
best_access_path().
This can grossly overestimate the number of rows expected during a join
and cause a bad join order.  This patch tries to infer a few simple
things about the derived table, like when the key being created matches
the group by list inside (if any) inside the derived table.

Approved by Sergei Petrunia (sergey@mariadb.com)
Tested by Lena Startseva (lena.startseva@mariadb.com)

Author: mariadb-RexJohnston

mysql-test/main/derived_cond_pushdown.result

@@ -4280,6 +4280,9 @@ INSERT INTO t1 VALUES
 (97,3290777,983,'2021-02-15 19:00:10'),(98,3290811,488,'2021-02-15 19:00:10'),
 (99,3290917,1283,'2021-02-15 19:00:36'),(100,3290858,482,'2021-02-15 19:00:42');
 
+insert into t1 select seq, 3300000+seq, 100+seq, '2021-02-09 18:31:35'
+from seq_101_to_1000;
+
 CREATE TABLE t2 (a int) ENGINE=MYISAM;
 INSERT INTO t2 VALUES
 (3289475),(3289496),(3289562),(3289593),(3289594),(3289595),(3289626),