fix: handle dynamic output changes while session is locked

When monitors are powered off (e.g. via niri power-off-monitors or
physical power switches), niri destroys and re-advertises the Wayland
outputs as they come back. Previously all three OutputHandler callbacks
were no-ops, causing two bugs:

- new_output: no lock surface was created for outputs that appeared
  after the initial lock, so the slowest monitor to wake up would show
  the compositor's red fallback instead of the lock screen.
- output_destroyed: stale LockedSurface, SessionLockSurface, output,
  and captured_background entries accumulated for gone outputs.

Fix new_output to create a lock surface (and register it with the lock
manager) whenever a new output appears while the session is locked.

Fix output_destroyed to remove the corresponding entries from
lock_surfaces, lock_manager.surfaces, outputs, and
captured_backgrounds, keeping all parallel vecs in sync. Add
LockManager::remove_surface_by_output to support this, returning the
removal index so lock_surfaces can be updated with the same index.

The all-monitors-off scenario (all outputs destroyed simultaneously)
is handled naturally: the vecs are emptied and repopulated as each
monitor fires new_output on wake.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/lock.rs

@@ -262,6 +262,17 @@ impl LockManager {
         action
     }
 
+    pub fn remove_surface_by_output(&mut self, output: &wl_output::WlOutput) -> Option<usize> {
+        use wayland_client::Proxy;
+        let output_id = Proxy::id(output);
+        let idx = self
+            .surfaces
+            .iter()
+            .position(|s| Proxy::id(s.output()) == output_id)?;
+        self.surfaces.remove(idx);
+        Some(idx)
+    }
+
     pub fn set_system_status(&mut self, status: SystemStatus) {
         for surface in &mut self.surfaces {
             surface.set_system_status(status.clone());