Improving GFlowNets for Text-to-Image Diffusion Alignment