M$^{2}$UGen: Multi-modal Music Understanding and Generation with the Power of Large Language Models