Feature request: Benchmark against tiny-nn

[oliver-batchelor]

I am guessing Dex is in no state for comparing against optimised implementations, but this seems like an ideal case for performance (and flexibility) can be gained over kernel-based array programming languages, especially if paired with AD for gradients. See the comparison of tensorflow vs. tiny-nn for small MLPs.

https://github.com/NVlabs/tiny-cuda-nn