On the Equivalence between Implicit and Explicit Neural Networks: A High-dimensional Viewpoint