On Interpretable Anomaly Detection Using Causal Algorithmic Recourse