Run safety moderation with Llama Guard 3 8B (@cf/meta/llama-guard-3-8b). POST messages or content to classify policy violations before downstream LLM calls.