Visualizing the Invisible: Generative Visual Grounding Empowers Universal EEG Understanding in MLLMs