REvolve: Reward Evolution with Large Language Models for Autonomous Driving