add param: doris_exchange_instances to set parallel after exchange

[xionglei0]

doris_exchange_instances speed up the query especially useful when:

1. data size decrease on a massive scale after exchange when using aggregation function or UDAF
2. cluster is large but we only need several instances to do aggregation. save computational resources

[xionglei0]

[kangkaisen]

@xionglei0 Hi, Thanks for your work!

1 Do you have the benchmark for this PR?

2 Could you use parallelExecInstanceNum instead of doris_exchange_instances?

[xionglei0]

"parallel_fragment_exec_instance_num" means the parallel during scan_node
"doris_exchange_instances" means the parallel after exchange_node
when scan datasize much more than datasize after exchange, like udaf : sum，count, doris_exchange_instances is useful

[xionglei0]

doris_exchange_instances value need to be set according to your data size and UDAF, we usually configure it when table is duplication model

[imay]

LGTM