Towards practical differentially private causal graph discovery