multigpu: Less constraints around framebuffers

[Drakulix]

The are a couple of commits that evolved around trying to use the cross_renderer with a GL-based and the pixman renderer.

(The biggest blocker for that is that the PixmanRenderer doesn't support it's own texture type as a framebuffer, because client shm WlBuffers. I have a commit "fixing" that, but it is ugly and I don't pursue this any more.)

Either way these commits are standalone and useful on their own. They can be reviewed one-by-one.

• 38ad389 - removes a left-over generic constraint, that does nothing
• 59998bc - Adds a way to access the target renderer. This is useful if you want to chain the initial render with some post-process effects in a generic way.
• 963a010 - This makes use of the fact, that if we don't have target in the MultiRenderer, we know that R == T (because the only way to get that is via cross_renderer, which always sets target). This we can have the Framebuffer-type only depend on T with some transmutes and remove the constraints on R. I am fairly certain this is correct and it doesn't crash, but I'd certainly appreciate a second look on this one.