CTP: Towards Vision-Language Continual Pretraining via Compatible Momentum Contrast and Topology Preservation