Improving Machine Translation with Large Language Models: A Preliminary Study with Cooperative Decoding