Multimodal contrastive learning for brain-machine fusion: From brain-in-the-loop modeling to brain-out-of-the-loop application.