If VLC is being destroyed for programmatic reasons (libvlc_release()),
the user might not have "quit".

Interfaces do not have an event handler that would justify the
separation. Indeed, no caller did anything in-between.

Also, the "primary" reference to the interface object belongs to the
main thread because of the libvlc cleanup procedure. Therefore, the
interface thread pointer really should not be returned to the creator.
Note that this does not really fix the small race condition but it
conceals it within intf_Create() and libvlc_InternalCleanup().

Only the core can safely destroy an interface, otherwise, the other
component could race against the core in joining the interface thread
or destroying the interface object.
The fire and forget status of interfaces is by design, and is expected
by -at least- the VLC binary, the NT service plugin and the Remote
Control interface plugin. Anyway, that is pretty the only point of
creating an interface instead of a private thread/object.

Conflicts:
	modules/video_output/drawable.c
	src/control/libvlc_internal.h

(see http://forum.videolan.org/viewtopic.php?f=32&t=59426)

This reverts commit 8018c67d.

It should return the current time even when the size isn't known.

Based upon an idea by Philipp Biermann.

An actual vlc-option for this will be created in the master branch.

Note that this feature does only affect VLCEmbeddedWindow - our plain vout window enforces the AR.

the need for compilation fixes with standard headers.

Clean up vcdx a bit. Or a lot. It compiles, but I can't test. Since apparently nobody uses this anyway it shouldn't be a big problem.

Refactor vcdx info.c dropping a few memory leaks and other problems. It compiles, but I can't test it.