Reasoning via Video: The First Evaluation of Video Models' Reasoning Abilities through Maze-Solving Tasks