fix(index): editor-exit deadlock — stop-aware waits on every game-thread dispatch

[whalemenace]

Repro: fresh install, let the first full index start, QUIT_EDITOR. You get "Indexing was still in progress during shutdown — force-stopped", DDC maintenance completes, then the process hangs forever — window gone, log stalled, kill required.

It's a cross-thread deadlock. The indexing worker parks in unbounded Event->Wait()s at fourteen sites, every one waiting on work only the game thread can run (throttle/critical/periodic GC via AsyncTask(GameThread), and every FMonolithCompilerSafeDispatch batch/sentinel dispatch, which needs a game-thread tick). Meanwhile Deinitialize() blocks the game thread in WaitForCompletion(). Neither side can ever proceed.

Changes:

• FIndexingTask::WaitForGameThreadEvent: all fourteen waits poll in 100ms slices and bail only when stop is requested AND IsEngineExitRequested() — a plain user cancel still waits normally, because the game thread is alive and will serve the task. Abandoned pooled events are deliberately leaked (a later Trigger on a leaked event is harmless; on a recycled one it's corruption).
• FMonolithCompilerSafeDispatch gains an opt-out abort token: an abandoned dispatch's Work payload — whose captures reference the unwound worker stack — is never invoked; the ticker triggers the event and unregisters.
• The registry-scan AsyncTask now owns its state via shared pointer instead of capturing the worker stack, so abandoning it is safe too.
• Deinitialize completes the still-Pending TaskNotification before destroying it. This assert (FSlateNotificationManager::ShutdownOnPreExit: "Missing call to SetComplete?") was always reachable on the force-stop path — the deadlock just hung the editor before it could fire.

Verified on 0.22.0: monolith_reindex force=true, quit mid-pass — the "force-stopped" warning fires and the editor exits in under 5 seconds, zero crash artifacts. Before the patch this hung 8+ minutes until killed.