DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledge