Do not add exchange when table's distributioin satisfy the distribution requirements

[liutang123]

issue Reason

Now, Doris will create two fragments for aggretion.
Sometime, Exchange for aggregation is unnecessary.
Think about follow cases:

1. aggregate an unpartitioned table.
   create table SQL:

CREATE TABLE `llj_test_1` (
  `dt` int(11) NOT NULL COMMENT "",
  `dis_key` varchar(20) NOT NULL COMMENT ""
) ENGINE=OLAP
DUPLICATE KEY(`dt`, `dis_key`)
COMMENT "OLAP"
DISTRIBUTED BY HASH(`dt`, `dis_key`) BUCKETS 3
PROPERTIES (
"replication_num" = "3",
"in_memory" = "false",
"storage_format" = "DEFAULT"
);

query SQL:select dt, dis_key,count(1) from llj_test_1 group by dt, dis_key;
2. aggregate a partitioned table.
create table SQL:

CREATE TABLE `llj_test` (
  `dt` int(11) NOT NULL COMMENT "",
  `dis_key` varchar(20) NOT NULL COMMENT ""
) ENGINE=OLAP
DUPLICATE KEY(`dt`, `dis_key`)
COMMENT "OLAP"
PARTITION BY RANGE(`dt`)
(PARTITION p20180822 VALUES [("19000101"), ("20181021")),
PARTITION p20181207 VALUES [("20181021"), ("20181022")))
DISTRIBUTED BY HASH(`dt`, `dis_key`) BUCKETS 3
PROPERTIES (
"replication_num" = "1",
"in_memory" = "false",
"storage_format" = "DEFAULT"
);

query SQL:select dt, dis_key,count(1) from llj_test group by dt, dis_key;

Suggestion

In DistributedPlanner, do not add the unnecessary Exchanges.
For case 1, we only need to judge that the table's distribute hash keys is a subset of the aggregate keys.
For case 2, we should jude two conditions:

• partition keys are also hash keys.
• the table's distribute hash keys is a subset of the aggregate keys.