Modin performance enhancements

[prutskov]

The table below contains a set of performance gaps, which take the most time part of execution in Modin performance tests.

Functionality section	Assignees	Issue	Comment	Priority	Status
merge operation	-	#4293	left join type	-	Not assigned
	-	#3588	inner join type	-	Not assigned
	-	#656	right/outer join type, defaults to pandas for now	-	Not assigned
groupby aggregation operations	@prutskov	#4288	groupby.mean reproducer is presented as example of aggregation operation. Applicable for flow DataFrameGroupBy._wrap_aggregation->qc.groupby_agg	-	✅
	-	#3585	groupby.mean	duplication	duplication
	-	#1901	groupby.mean	duplication	duplication
getitem operations	@dchigarev	#4268	__getitem__ operation, _get_dict_of_block_index() is hotspot	-	✅
	@dchigarev	#1903	iloc, _get_dict_of_block_index() is hotspot	duplication	duplication
binary operations	@prutskov	#4182	operands are Modin DataFrame/Series	-	✅
	@prutskov	#3100	operands are Modin DataFrame/Series, a lot of details (need to verify after #4391)	-	WIP
drop operation	@dchigarev, @prutskov	#3844	3 bottlenecks are in API, QC layers and _get_dict_of_block_index()	-	✅
sort_values operation	@RehanSD	#3535	defaults to pandas for now	-	@RehanSD
read_parquet operation	@pyrito	#4305	Only parallelization between columns/column-partitions is supported for now	-	✅
constructing DataFrame from dict with Modin Series values	-	#4263	flow is from_non_pandas	-	Not assigned
	-	#1572	According with the flow description this is duplication	duplication	duplication

[YarShev]

Closing this issue as outdated. We will track the unresolved issues separately.