The actual window geometry should stay in the same position
regardless of surface geometry being changed.

We're interested in restoring the actual window into the same
position, regardless of whether the shadows around it happened
to change their size meanwhile.

This makes it consistent with PHOC_EDGE_SNAP_THRESHOLD and actually makes
more sense.

This follows the convention used by other projects like phosh.

We've been using proper wlroots packaging for a long time now.

This allows them to run as soon as the dependency is fulfilled, before
the previous stage has been completed.

This ensures that the view is still on the list while it may get
damaged by children handlers - otherwise phoc_desktop_view_is_visible
won't be able to work correctly.

Fixes #191

If the view is unmapped, we can tell that it's not visible right
away. Later assertion assumes that there's at least one view on
the desktop->views list, which is a valid assumption to make only
if the view we got is actually mapped.

Should fix #191

Every wlr_output is going to have an associated PhocOutput, but
wlr_output_layout_output_at is not guaranteed to always return
a non-NULL value. Therefore it doesn't make much sense to check
for `output` being NULL right after unchecked dereference
of `wlr_output`.

It's useful to have when testing custom wlroots builds on the phone.

We still want to be able to reveal shell when there's no surface
at the edge of the screen - which can happen when the fullscreen
surface's size doesn't cover the whole output.

Some apps open windows with preferred size that are larger than
the phone screen, but that can be resized down manually to fit.
There's no reason not to try to make them fit on the screen
automatically, so do just that when centering views.

It's simpler and does the same thing :)

It's not really a useful feature to have in a mobile compositor and just
makes positioning code more complicated for no good reason. It wasn't
even complete, as evidenced by TODOs being removed with it. Support for
iterating over and drawing rotated surfaces is retained as it may still
be useful, but we sure don't need it at the view level.

Setting force_shell_reveal changes the contents of the output, so
it needs to be damaged. It is already done in all the other places
where it's changed.

For consistency.

So far phoc behaved like a regular stacking window manager, putting
each focused toplevel surface on top of all the others. This, combined
with lack of sophisticated scene graph, resulted in all of the surfaces
being redrawn each time there was a damage in part of the output that
those surfaces cover. In automaximize mode, this means that we always
redraw all of the windows regardless of whether they're visible or not.

However, when we have a single output and we're in automaximize mode
(so, most of the time phosh is in use), we can stop redrawing unnecessary
surfaces pretty easily with bunch of simple visibility checks.

This commit introduces a way for render code to check visibility and adds
bunch of such checks. In the future, those checks could even be replaced
with a proper scene graph-based visibility checking that takes opaque
regions into account.

"Restoring" means putting the window back into unmaximized
and untiled state.

This mirrors view_arrange_maximized and allows to rearrange the view
when necessary without influencing its state.

We only want unmaximized and untiled state to be stored, since restoring
normal state is what this state backup is getting used for.

This codepath is useful for tiled surfaces as well.

This is not an error - it's just an action that doesn't trigger any
special behavior. There's no need to even log it.

Without that, we're left with special cursor until the pointer
position gets updated again.

It's not a complete fix yet because it doesn't handle surface geometry, but
it's at least much better than what it used to be.

This prevents the new position from being overriden with an older
one that was set by view_move_resize and was waiting for the resize
to happen client-size before being applied.

Seems like some old artifact from rootston, which even got removed
in https://github.com/emersion/rootston/commit/df040786a70f24d7c112c13bfde5249db49624bb