Tiny-Align: Bridging Automatic Speech Recognition and Large Language Model on the Edge