Grad should be a Value instead of python/numpy scalar

[MarisaKirisame]

So you can do higher order autodiff.

[karpathy]

yes, similar to PyTorch, a good feature request. I'll see if I can find a way to do it without sacrificing the clarity, cleanliness and compactness.

[evcu]

I played with higher-order grads a bit in my fork. It was this plus some small changes. Created a pr: #8