Evaluate the pros and cons of lazy functions submission (via partition.add_to_apply_calls)

[dchigarev]

Do we want to make this lazy? Since split_row_partitions is in effect the properly partitioned dataframe, we can transform to col partitions, and then add_to_apply_calls the sort instead, and defer metadata materialization till it's needed?

Originally posted by @RehanSD in #5780 (comment)

Modin's partition class has a feature that seems to not be used at all. The feature allows forming a queue of remote calls making it possible to combine them as a single remote call to the execution engine when the materialization is actually needed, thus reducing the number of remote calls being made to the engine and potentially optimizing the workflow:

modin/modin/core/dataframe/pandas/partitioning/partition.py

Lines 117 to 119 in 94b2a0e

	def add_to_apply_calls(self, func, *args, length=None, width=None, **kwargs):
	"""
	Add a function to the call queue.

We once had an attempt to use this feature (#1838) which however resulted in performance regression for certain cases and thus the changes were reverted by #2471. We never tried to re-iterate the changes and figure out the root cause of the regression.

So opening this issue as a reminder that we may want to do so as IMO it has a great potential for optimizing the way we operate with the execution engine.

[YarShev]

Related to #6731.