AdversaBench: Automated LLM Red-Teaming with Multi-Judge Confirmation and Cross-Model Transferability