Optimization for distributed GROUP BY with ORDER BY without aggregate functions.

[alexey-milovidov]

Use case
If the query contains GROUP BY and either does not contain ORDER BY or contains ORDER BY that does not depend on aggregate functions and only possibly depend on aggregation keys and does not depend on non-deterministic functions or there is no aggregate functions, we should apply ORDER BY transformation with LIMIT on remote servers. If there is no ORDER BY but aggregate functions are present, add implicit ORDER BY keys.

This will help to send less amount of data over network and parallelize ORDER BY.