Hyper-SD is a trajectory segmentation consistency model for efficient image synthesis, supporting 1 to 8-step reasoning with human feedback learning and score distillation