Approximate GROUP BY with ORDER BY ... LIMIT distributed query.

When doing distributed query with

GROUP BY ... ORDER BY ... LIMIT ...

we can get approximate result if perform ORDER BY and LIMIT on every shard and then merge filtered data on the initiating node.

It will send less data over network making query tremendously faster.
The results can be accurate enough and it will be useful to make this behaviour available under a setting.

The result will be accurate if top keys appeared to be the same on every shard.
We can also allow to tune the accuracy by multiplying limit to some value.

See also:
\- `max_rows_to_group_by` and `group_by_overflow_mode = 'any'`;