It’s DONE: Direct ONE-shot learning with Hebbian weight imprinting