Sql. Improve intra-query parallelism

At the moment, intra-query parallelism is achieved by splitting the query on several fragments in the place where distribution conversion is required, but the fragment itself is tied up to a single thread. Such a parallelization results in a low resource utilization by a single query, especially in case of huge fragments like the ones containing aggregation or colocated joins.

Let's investigate possible solutions and make a POC