Optimizing molecular geometry using reinforcement learning with density functional theory