Learning and Leveraging World Models in Visual Representation Learning