nstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generation