Can LLM Agents Infer World Models? Evidence from Agentic Automata Learning