Beyond Independent Frames: Latent Attention Masked Autoencoders for Multi-View Echocardiography - Databubble