VRoPE: Rotary Position Embedding for Video Large Language Models