A Multi-Task Foundation Model for Wireless Channel Representation Using Contrastive and Masked Autoencoder Learning