OledFL: Unleashing the Potential of Decentralized Federated Learning via Opposite Lookahead Enhancement