ManagerTower: Aggregating the Insights of Uni-Modal Experts for Vision-Language Representation Learning