Efficient burn-back prediction and reverse design of solid rocket motor star grains using deep operator learning