ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments