By Fair Means or Foul: Quantifying Collusion in a Market Simulation with Deep Reinforcement Learning