Integrally Migrating Pre-trained Transformer Encoder-decoders for Visual Object Detection