Samsung Researchers Introduce LoRA-Guard: A Parameter-Efficient Guardrail Adaptation Method that Relies on Knowledge Sharing between LLMs and Guardrail Models