Credit assignment & meta-learning in a single lifelong trial: the deepest kind of deep RL