Improve performance of distinct aggregations

[Dith3r]

Description

Improve performance of distinct aggregations by defining additional strategies based on source properties (for example, NDV).

The strategy to use for multiple distinct aggregations.
SINGLE_STEP Computes distinct aggregations in single-step without any pre-aggregations.
This strategy will perform poorly if the number of distinct grouping keys is small.
MARK_DISTINCT uses MarkDistinct for multiple distinct aggregations
or for mix of distinct and non-distinct aggregations.
PRE_AGGREGATE Computes distinct aggregations using a combination of aggregation
and pre-aggregation steps.
AUTOMATIC chooses the strategy automatically.

Single-step strategy is preferred. However, for cases with limited concurrency due to
a small number of distinct grouping keys, it will choose an alternative strategy
based on input data statistics.

Strategy	Duration	Query
MARK_DISTINCT	5353ms	select count(ss_customer_sk), count(distinct ss_ticket_number) from hive.tpcds_sf1000_orc.store_sales;
PRE_AGGREGATE	2468ms	select count(ss_customer_sk), count(distinct ss_ticket_number) from hive.tpcds_sf1000_orc.store_sales;
MARK_DISTINCT	10109ms	select count(ss_quantity), count(distinct ss_item_sk), count(distinct ss_store_sk) from hive.tpcds_sf1000_orc.store_sales;
PRE_AGGREGATE	8253ms	select count(ss_quantity), count(distinct ss_item_sk), count(distinct ss_store_sk) from hive.tpcds_sf1000_orc.store_sales;

Additional context and related issues

Release notes

( ) This is not user-visible or is docs only, and no release notes are required.
( ) Release notes are required. Please propose a release note for me.
(x) Release notes are required, with the following suggested text:

# General
* Improve performance of distinct aggregations. ({issue}`21907`)
  Replaced configuration property `optimizer.mark-distinct-strategy` with `optimizer.distinct-aggregations-strategy`. 
  It's allowed values are changed as follows: `NONE -> SINGLE_STEP` and `ALWAYS -> MARK_DISTINCT`. 
  Replaced `optimizer.optimize-mixed-distinct-aggregations` configuration property 
  with a new `PRE_AGGREGATE` option in `optimizer.distinct-aggregations-strategy` .

[lukasz-stec]

lgtm % comments