A data scientist is running dozens of experiments across multiple model families (XGBoost, LightGBM, and scikit-learn RandomForest) within a single MLflow experiment. Later, they want to be able to easily filter and group runs by the model family used, but 'model_family' is not something MLflow captures automatically as a parameter or metric. What is the recommended way to annotate each run so it can be queried later with a filter like tags.model_family = 'xgboost'?