Exploring the Effect of Reinforcement Learning on Video Understanding: Insights from SEED-Bench-R1