Enhancing Text-to-Image Diffusion Transformer via Split-Text Conditioning