NetHack - What a game from 1987 tells us about the current limits of Deep Reinforcement Learning