Strategic Intelligence in Large Language Models: Evidence from Evolutionary GT