Online Adaptive Policy Selection in Time-Varying Systems: No-Regret via Contractive Perturbations