FreGAN: Exploiting Frequency Components for Training GANs under Limited Data