COHESIV: Contrastive Object and Hand Embeddings for Segmentation In Video