Bridging the Gap: Regularized Reinforcement Learning for Improved Classical Motion Planning with Safety Modules