Multi-Modal Continual Test-Time Adaptation for 3D Semantic Segmentation