Multi-agent online learning in time-varying games