Revisiting Deepfake Detection: Chronological Continual Learning and the Limits of Generalization