VLMAE: Vision-Language Masked Autoencoder