Building Bridges, Not Walls -- Advancing Interpretability by Unifying Feature, Data, and Model Component Attribution