Steering No-Regret Agents in MFGs under Model Uncertainty