Follow up of D8418.
Optionally moves the whole texture upload into a separate thread using a shared GL context. On OS X this allows to use glFinish() after each texture upload without having to block the main thread.
This is still something I'm experimenting on and testing on, nothing final. But still, it's working nicely and looks very promising so far - no regressions. It even works under corrosive test conditions (i.e. using Giphy which breaks the current top of tree).