Rethinking animal attrition in preclinical research: expressing causal mechanisms of selection bias using directed acyclic graphs