Function mnist_mlp 

pub fn mnist_mlp<D: Float>() -> impl Fn(SymTensor) -> SymTensor

Fully-connected MLP for MNIST — no conv layers, suitable for end-to-end training.

Input         [N,  1, 28, 28]
Flatten       →  [N, 784]
Linear(784→256, no bias)
ReLU
Linear(256→64, no bias)
ReLU
Linear(64→10,  no bias)   →  [N, 10]  raw logits (no softmax)