Is Causal Abstraction Enough For Mechanistic Interpretability?