Add an official guide (and boilerplate) for adding new optimized features to NumDot for custom builds.

[Ivorforce]

If somebody really needs the best performance from their math, they should be able to use xtensor directly. NumDot will be there to provide 'standard' functionality and the framework for how to compile it directly, and to all platforms.

The best way to do this might be to allow adding NumDot as a submodule that is embedded in another GDExtension. Then, it can truly be integrated into whatever the game needs.