In the context of developing a workflow that involves training multiple models and comparing their performance, which MLflow feature is most suitable for logging metrics and visualizations for each model run?