Adapting Pretrained Text-to-Text Models for Long Text Sequences