Toward a Fundamental Model for Sleep Analysis with a Multimodal Hybrid Self-Supervised Learning Framework