SignMouth: Leveraging Mouthing Cues for Sign Language Translation by Multimodal Contrastive Fusion