Run cores into a subprocess

Libretro cores should be run into a subprocess, this would ultimately allow to:

• simplify identifying the calling-back core, hence
• support cores calling back from a worker thread,
• make retro-gtk more resilient to crashes in the Libretro core,
• optionally use libseccomp to mitigate attacks coming from malicious cores or malicious games.

To do so, I propose to split RetroCore into a stub living in the user's process and a skeleton living in a worker process.