VERIRL: Boosting the LLM-based Verilog Code Generation via Reinforcement Learning