Harmonizing Supervised Fine-Tuning and Reinforcement Learning with Reward-Based Sampling for Continual Machine Unlearning | Synapse