core: Avoid queueing resizes while changing workspace during window drag
We already have a user move/resize going on, so avoid botching it with
window resizes with flags & META_MOVE_RESIZE_USER_ACTION == 0 that will
trigger unintended constraints.
wayland: Detach MetaWaylandActorSurface from its actor on dispose
And ensure the actor is no longer reactive even though it might live longer
because of close effects, GCs, and whatnot. This ensures the actor is not
eligible for pointer picking within the destruction of its surface.
The MetaCloseDialog implementation object may stay artifically alive
for a longer period. This was usually fine till gnome-shell commit
b03bcc85aad, as the check_alive() timeout will keep running even
though the window went unmanaged/destroyed, leading to crashes.
In order to fix this, forcibly hide the dialog if it is visible and
the window is being unmanaged, so the timeout is stopped in time.
wayland: Handle get_subsurface() with a role-less parent surface
The order of role creation is undetermined, so we can't account that
the parent surface will have a role (and an actor) at the time of
creating the wl_subsurface role for a child surface.
So we must do it both ways, add the subsurface as a child on
get_subsurface() if the parent already got a role, and lazily add
child subsurface actors to the current one if the parent surface got
it at a later point.
renderer/native: Also wrap flip closures for EGLStreams
When using the EGLStream backend, the MetaRendererNative passed a
GClosure to KMS when using EGLStreams, but KMS flip callback event
handler in meta-gpu-kms.c expected a closure wrapped in a closure
container, meaning it'd instead crash when using EGLStreams. Make the
flip handler get what it expects also when using EGLStreams by wrapping
the flip closure in the container before handing it over to EGL.
`modelview` is uninitialized and the `apply` function just multiplies it.
What we really want is to initialize `modelview` so replace `apply` with
`get`.
