SEA-SafeguardBench: Evaluating AI Safety in SEA Languages and Cultures