Q-ROU: Quantization-Robust Orthogonal Unlearning via Active Retention for Post-Deployment Knowledge Removal in Large Language Models | Synapse