Multimodal Continual Learning with MLLMs from Multi-scenario Perspectives