Can Large Language Models Develop Strategic Reasoning? Post-training Insights from Learning Chess