planner should support 3 stage aggregation for single scalar distinct agg

[fixdb]

Enhancement

Currently, count(distinct) in MPP mode still execute in a single worker, which is bad when there are large number of distinct values. e.g.:

mysql> explain select count(distinct L_ORDERKEY) from lineitem;
+------------------------------------+----------+--------------+----------------+-----------------------------------------------------------+
| id                                 | estRows  | task         | access object  | operator info                                             |
+------------------------------------+----------+--------------+----------------+-----------------------------------------------------------+
| TableReader_18                     | 1.00     | root         |                | data:ExchangeSender_17                                    |
| └─ExchangeSender_17                | 1.00     | mpp[tiflash] |                | ExchangeType: PassThrough                                 |
|   └─Projection_13                  | 1.00     | mpp[tiflash] |                | Column#18                                                 |
|     └─HashAgg_14                   | 1.00     | mpp[tiflash] |                | funcs:count(distinct test.lineitem.l_orderkey)->Column#18 |
|       └─ExchangeReceiver_16        | 1.00     | mpp[tiflash] |                |                                                           |
|         └─ExchangeSender_15        | 1.00     | mpp[tiflash] |                | ExchangeType: PassThrough                                 |
|           └─HashAgg_6              | 1.00     | mpp[tiflash] |                | group by:test.lineitem.l_orderkey,                        |
|             └─TableFullScan_12     | 10000.00 | mpp[tiflash] | table:lineitem | keep order:false, stats:pseudo                            |
+------------------------------------+----------+--------------+----------------+-----------------------------------------------------------+
8 rows in set (0.01 sec)

We should be able to generate a plan with 3 stage aggregation for scalar distinct agg.
NB. only for cases where there is only 1 distinct agg function.