fix(sharednd): address review — stop/start race, onExit cleanup, no busy loop off niri

- Debounce stop transitions by 1 s (host update tick): a stop/start pair
  from switching casts collapses into no transition, so the async
  `dnd-set off` can no longer race the `dnd-status` query and drop
  ownership mid-share. Also stops DND flapping.
- Add onExit: if the plugin owns DND when disabled/reloaded, turn it off
  via a detached runAsync (survives VM teardown).
- Gate detection on NIRI_SOCKET + commandExists("niri") instead of an
  unconditional retry loop: on other compositors (or niri installed but
  not running) the service now spawns nothing. If NIRI_SOCKET is set but
  the binary is missing from PATH, warn once.
- Make the stream retry loop self-terminating (parent gone or niri
  socket removed): hard host exits skip process-group cleanup and were
  leaking orphaned loops that respawned `niri msg` every 3 s across
  sessions (observed 11 such orphans over 4 days).
- Bump version to 1.1.0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

sharednd/plugin.toml

@@ -5,13 +5,15 @@
 # appears, notification DND is enabled through the host IPC
 # (notification-dnd-set); when the last one disappears, DND is restored.
 # If DND was already on before sharing started, the plugin does not take
-# ownership and leaves it on afterwards (configurable).
+# ownership and leaves it on afterwards (configurable). Detection only
+# starts inside a niri session (NIRI_SOCKET + niri binary present); on
+# other compositors the service is inert and spawns nothing.
 #
 # External requirements: niri (screencast events come from niri IPC).
 
 id = "whyoolw/sharednd"
 name = "ShareDND"
-version = "1.0.0"
+version = "1.1.0"
 min_noctalia = "5.0.0"
 author = "whyoolw"
 license = "MIT"