Gravar-mail: On the Relevance of the Loss Function in the Agatston Score Regression from Non-ECG Gated CT Scans