Parameterized Action Reinforcement Learning for Inverted Index Match Plan Generation