Towards Assurance of LLM Adversarial Robustness using Ontology-Driven Argumentation