Sam 3: a unified foundation model for promptable segmentation in images/videos