Planner push down TopN through IndexMerge

[fixdb]

Enhancement

create table foo(a int primary key, b int, c int, d int, e int);
create index index_bda on foo(b,d,a);
create index index_cda on foo(c,d,a);

explain select * from foo where b=10 or c=10 order by d desc limit 10;
+----------------------------------+---------+-----------+-------------------------------------+-----------------------------------------------+
| id                               | estRows | task      | access object                       | operator info                                 |
+----------------------------------+---------+-----------+-------------------------------------+-----------------------------------------------+
| TopN_10                          | 10.00   | root      |                                     | test.foo.d:desc, offset:0, count:10           |
| └─IndexMerge_25                  | 19.99   | root      |                                     | type: union                                   |
|   ├─IndexRangeScan_22(Build)     | 10.00   | cop[tikv] | table:foo, index:index_bda(b, d, a) | range:[10,10], keep order:false, stats:pseudo |
|   ├─IndexRangeScan_23(Build)     | 10.00   | cop[tikv] | table:foo, index:index_cda(c, d, a) | range:[10,10], keep order:false, stats:pseudo |
|   └─TableRowIDScan_24(Probe)     | 19.99   | cop[tikv] | table:foo                           | keep order:false, stats:pseudo                |
+----------------------------------+---------+-----------+-------------------------------------+-----------------------------------------------+
5 rows in set (0.01 sec)

Without IndexMerge, we can push TopN down to TiKV:

explain select * from foo where b=10 order by d desc limit 10;
+--------------------------------+---------+-----------+-------------------------------------+----------------------------------------------------+
| id                             | estRows | task      | access object                       | operator info                                      |
+--------------------------------+---------+-----------+-------------------------------------+----------------------------------------------------+
| IndexLookUp_24                 | 10.00   | root      |                                     | limit embedded(offset:0, count:10)                 |
| ├─Limit_23(Build)              | 10.00   | cop[tikv] |                                     | offset:0, count:10                                 |
| │ └─IndexRangeScan_21          | 10.00   | cop[tikv] | table:foo, index:index_bda(b, d, a) | range:[10,10], keep order:true, desc, stats:pseudo |
| └─TableRowIDScan_22(Probe)     | 10.00   | cop[tikv] | table:foo                           | keep order:false, stats:pseudo                     |
+--------------------------------+---------+-----------+-------------------------------------+----------------------------------------------------+
4 rows in set (0.00 sec)

[winoros]

The basic needs of this issue have been solved. Closed this one.
The later improvements will be tracked by other issues.