Tricks or Traps? A Deep Dive into RL for LLM Reasoning