Learning Reach-Avoid Task with Reinforcement Learning: Vectorized Simulation and Benchmark