Scaling Laws for Imitation Learning in NetHack