Add CUDA code for sort_vertices.

[dfalbel]

This PR adds changes to lantern so we can add custom CUDA code.
We added the code from https://github.com/lilanxiao/Rotated_IoU use to efficiently sort vertices.

This will eventually fix #608

TODO:

• Create the Rcpp wrapper around the lantern call.
• Deal with the case when lantern is not built with cuda support. Ie, we need to provide an implementation of sort_vertices that works in that case.
• Add comments attributing to the original repository + license.

[dominikj2]

Hi @dfalbel,

Thank you for PR that adds changes to lantern so we can add custom CUDA code.

Reinstalled torch and found torch::contrib_sort_vertices() . Thank you!

[dfalbel]

@dominikj2 Glad it worked! Let me know if you have any problems. It would be nice if at some point we provided a cpu implementation for contrib_sort_vertices(). Currently it will only work if all inputs are on the GPU.