Nerve theorems for fixed points of neural networks