v.25.4Performance Improvement

Now we use number of replicas

Now we use number of replicas to determine task size for reading with parallel replicas enabled. This provides better work distribution between replicas when the amount of data to read is not really big. #78695 (Nikita Taranov).