Boosting Alignment for Post-Unlearning Text-to-Image Generative Models