Log model evaluation metrics to central results sheet

Your forecasts arrive without evaluation, so engineers lack accuracy context for workshops. It logs RMSE and high-error flags to a shared sheet so ML engineers get workshop-ready metrics quickly.