Where to go next: Learning a Subgoal Recommendation Policy for Navigation in Dynamic Environments