Towards zero-forget continual learning for interactive trajectory prediction: a dynamically expandable approach | Synapse