CrossDTR: Cross-view and Depth-guided Transformers for 3D Object Detection