Killed-State Call Lifecycle Fixes

We have successfully resolved the issues surrounding incoming calls received while the app is in a killed (quit) state. The solution ensures a seamless transition from answering a native notification to speaking in the Agora CallRoomScreen, with proper cleanup of native UI and navigation stacks.

🚀 Key Issues Resolved

1. Headless Hook & API Failures

• Problem: When the app was killed, the headless JS context attempted to confirm the "Answer" action via acceptCall. However, CallApi required React context (Providers), which didn't exist yet, causing the API to fail and the caller to drop the call after a few seconds.
• Solution: Implemented a Headless Fallback in CallApi.js. It now detects when React isn't ready and uses apiClient directly, fetching the user's authentication token from AsyncStorage to notify the backend immediately.

2. The "Second Popup" & Setup Cache

• Problem: RNCallKeep setup often fails in headless mode because no Android Activity exists. Our setupPromise was caching this failure, causing all subsequent retry attempts in the main app to fail instantly without actually retrying.
• Solution:
  • Updated setupCallKeep to reset its promise on failure, allowing the main app to successfully retry.
  • Added an exponential backoff retry loop (500ms, 1500ms, 3000ms) during app initialization.
  • Added a best-effort direct native UI dismissal in the headless context.

3. Call Dropping due to Unmounting

• Problem: During cold boot, the Deep Link would open CallRoomScreen. In parallel, SplashScreen would finish loading and execute navigation.replace(SCREEN.DRAWER). This swallowed the CallRoomScreen, unmounting it and triggering Agora's leaveChannel, which killed the call.
• Solution: Refactored SplashScreen.js to detect if a call has already been navigated to. If so, it now returns early and does nothing, allowing the CallRoomScreen to remain safely on top of the stack.

4. Navigation Stack Stranding

• Problem: After a cold-boot call ended, navigation.goBack() would take the user back to the SplashScreen, which was frozen because it had skipped its transition to the Home screen.
• Solution: Upgraded safeGoBack in CallRoomScreen.js with Stack Introspection. It now checks the navigation state before popping; if the screen underneath is the Splash screen or the stack is empty, it resets the entire root to the DrawerNavigator.

5. Deep Link Logic Safety

• Problem: Native "End Call" events were firing deep links that App.js was interpreting as requests to navigate to the call screen.
• Solution: Modified handleDeepLink in App.js to explicitly check for the action=accept parameter. Action strings for end are now ignored for navigation purposes.

📱 Verification Results

Scenario	Behavior	Result
App Killed -> Receive Call -> Accept	UI opens directly to CallRoom; API confirms instantly; Caller stays connected.	✅ Fixed
Call Room Active -> Caller Ends	CallRoom closes and user is redirected to Home Drawer (not Splash).	✅ Fixed
Stale Native UI	Persistent notifications are dismissed automatically within 1-2 seconds of app boot.	✅ Fixed
UI Reset Flash	SplashScreen transition is bypassed during active calls, preventing the UI "flicker".	✅ Fixed