Readback results from HVM into Rust through interop

[edusporto]

This allows the resulting nets to be read back directly from Rust, without requiring each implementation to print it's networks in error-prone ways.

TODO: The CUDA version is still quite slow due to copying the entire GNet into the host from the device. The way to fix it is to compact the global network's memory and only copy the relevant portion before copying it into RAM.