qsvenc: take only the allocated dimensions from the frames context

Other parameters, like the display size, should still be taken from the
codec context.