Drop extra parallel_reduce() when sorting objects in distributed tree

Created by: dalg24

Following up on https://github.com/arborx/ArborX/pull/248#issuecomment-605071389