ArtAug: Enhancing Text-to-Image Generation through Synthesis-Understanding Interaction