Improving LLM Reasoning through Scaling Inference Computation with Collaborative Verification