Modeling Discrimination with Causal Abstraction