CoConGAN: Cooperative contrastive learning for few-shot cross-domain heterogeneous face translation