Enhancing Low-Rank Adaptation with Recoverability-Based Reinforcement Pruning for Object Counting Environment python=3.8 DRL+SAM+JZ