HyperDAS: Towards Automating Mechanistic Interpretability with Hypernetworks