Selective Progress-Aware Querying for Human-in-the-Loop Reinforcement Learning