A Counterfactual Safety Margin Perspective on the Scoring of Autonomous Vehicles' Riskiness