The House Always Wins: A Framework for Evaluating Strategic Deception in LLMs