feat: fix critical bugs, optimize AppBlockerService battery usage, and add AI validation selector

[Afrasyaab-GH]

Hi there! This Pull Request resolves several critical bugs in the navigation and sync layers, optimizes background service performance to prevent battery drain, resolves Android 14+ foreground service crashes, and introduces a unified AI Validation Engine settings configuration screen.

---

🛠️ Changes Made

1. Android 14+ Foreground Service Crash Fixes (New)

• AppBlockerService Type Update: Changed the service type from dataSync to specialUse (with subtype app_blocking_and_screentime_restrictions in the manifest). This prevents the strict 24-hour runtime exhaustion limit on Android 14+ (which was throwing ForegroundServiceStartNotAllowedException).
• Background Start Safety: Wrapped all startForegroundService invocations in try-catch blocks across MainActivity, DeepFocusQuestView, OnboardingScreen, and AppListViewModel to safely handle background start restrictions without crashing.

2. Navigation & Sync Bug Fixes

• Route Typo in RootRoute.kt: Corrected ShowTutorials route string from "showTutorial{/" to "showTutorial/" to prevent navigation crashes.
• MainActivity Sync Parallelization: Wrapped unSyncedQuestItems and unSyncedStatsItems collections in separate launch coroutine builders so they execute in parallel. Stats are now successfully synced without getting blocked by the quest flow.
• Database Loop in StatsSyncWorker.kt: Changed the worker to mark stats as synced on the statsRepository instead of the questRepository, resolving the infinite synchronization loop.
• FCM Gift Coins Key in FcmHandler.kt: Corrected the lookup key from "coins" to "gift_coins".
• Boost Activation in UserRepository.kt: Updated activateBoost to store the actual passed item instead of hardcoding InventoryItem.XP_BOOSTER.
• Screentime CoerceAtLeast Fix: Refactored ScreenUsageStatsHelper.kt and UsageStatsHelper.kt to prevent start time constraints from being silently discarded.

3. AppBlockerService battery & UI optimizations

• Refactored AppBlockerService.kt to run the 100ms polling loop on a dedicated background thread (HandlerThread "AppBlockerMonitor") instead of the main/UI thread.
• Removed high-frequency synchronous disk I/O (loadUnlockedAppsFromServer()) from the tight 100ms loop and set up cache updates reactively.

4. AI Validation Selector Settings & Failure Recovery UI

• Settings Selection UI: Added a settings selector card in the profile page so users can toggle validation between QuestPhone Cloud Server, Local On-Device AI, and Private Gemini API Key.
• Enhanced Failure Recovery Screen: If a validation fails, the UI displays a radio-group selector enabling you to switch engines immediately. It includes inline API key input and lets you download/select ONNX models directly without returning to settings.
• Automatic Engine Selection: Downloading/selecting a model in ModelDownloadDialog now automatically switches your active engine to "local".

---

⚠️ Notes on Testing & Validation Engines

• Private Gemini API Key Mode: Thoroughly tested and works perfectly. Captures are successfully sent to Gemini and correctly validated.
• QuestPhone Cloud Server Mode: Untested (I have not tested the default server/cloud-based validation engine).
• Local On-Device AI Mode (ONNX/SIGLIP): There are known issues with the local HuggingFace SIGLIP model files parsing (throwing ORT_INVALID_PROTOBUF errors during load time on ONNX Runtime), which remain unresolved. Selecting local models will show a description dialog but may fail during evaluation.

[nethical6]

tchhh, you could've just made a custom quest for adding that nano gemini support

[nethical6]

i'd like to merge this but its been so long i last worked on this project that i don't remember a thing. so it will take a bit more time.

[Afrasyaab-GH]

i'd like to merge this but its been so long i last worked on this project that i don't remember a thing. so it will take a bit more time.

No worries at all, please take all the time you need to review it! There is absolutely no rush.

tchhh, you could've just made a custom quest for adding that nano gemini support
Haha!

Regarding the Nano support, a custom quest is definitely a cool alternative! We ended up building it directly into the AI Snap Quest pipeline so that users can seamlessly toggle between Cloud, Private API Key, and Local Offline (Nano/ONNX) validation engines depending on their device capabilities and privacy preferences.

I also just pushed a few updates to address some critical crashes and UI bugs:

1. Android 14+ Foreground Service Crash Fixes: Switched AppBlockerService from dataSync to specialUse type to bypass the strict 24-hour dataSync accumulated runtime limit on Android 14+. I also wrapped the service start calls in try-catch blocks to handle background launch restrictions without crashing the app.
2. Inline Recovery UI: If validation fails, the evaluation dialog now displays a direct selection menu where you can configure/save your Gemini API key or change engines inline without having to exit back to Settings.
3. Photo Preview Fix: Corrected the evaluation screen's photo path to pull from filesDir (matching where CameraScreen saves it) so the captured snapshot preview renders correctly on screen during validation.

[Afrasyaab-GH]

⚠️ Notes on Testing & Validation Engines

• Private Gemini API Key Mode: Thoroughly tested and works perfectly. Captures are successfully sent to Gemini and correctly validated.
• QuestPhone Cloud Server Mode: Untested (I have not tested the default server/cloud-based validation engine).
• Local On-Device AI Mode (ONNX/SIGLIP): There are known issues with the local HuggingFace SIGLIP model files parsing (throwing ORT_INVALID_PROTOBUF errors during load time on ONNX Runtime), which remain unresolved. Selecting local models will show a description dialog but may fail during evaluation.

[Afrasyaab-GH]

PLease someone test the

QuestPhone Cloud Server Mode: Untested (I have not tested the default server/cloud-based validation engine).
Local On-Device AI Mode (ONNX/SIGLIP): There are known issues with the local HuggingFace SIGLIP model files parsing (throwing ORT_INVALID_PROTOBUF errors during load time on ONNX Runtime), which remain unresolved. Selecting local models will show a description dialog but may fail during evaluation.

I will try to fix it but I can not promise