Fix flickering issues due to reconnecting loop

lib/providers/stream_recovery.dart

@@ -109,6 +109,7 @@ class StreamRecoveryWrapper<T> {
   int _recoveryAttempts = 0;
   Timer? _pollingTimer;
   Timer? _recoveryTimer;
+  Timer? _stabilityTimer;
   StreamSubscription<List<Map<String, dynamic>>>? _realtimeSub;
   StreamController<StreamRecoveryResult<T>>? _controller;
   bool _disposed = false;
@@ -159,8 +160,28 @@ class StreamRecoveryWrapper<T> {
 
   void _onRealtimeData(List<Map<String, dynamic>> rows) {
     if (_disposed) return;
-    // Successful data — reset recovery counters.
-    _recoveryAttempts = 0;
+
+    // When recovering, don't reset _recoveryAttempts immediately.
+    // Supabase streams emit an initial REST fetch before the realtime
+    // channel is established.  If the channel keeps failing, resetting
+    // on that REST data creates an infinite loop (data → reset → subscribe
+    // → fail → data → reset …).  Instead, start a stability timer — only
+    // reset after staying connected without errors for 30 seconds.
+    if (_recoveryAttempts > 0) {
+      _stabilityTimer?.cancel();
+      _stabilityTimer = Timer(const Duration(seconds: 30), () {
+        if (!_disposed) {
+          _recoveryAttempts = 0;
+          debugPrint(
+            'StreamRecoveryWrapper[$channelName]: '
+            'connection stable for 30s — recovery counter reset',
+          );
+        }
+      });
+    } else {
+      _recoveryAttempts = 0;
+    }
+
     _setStatus(StreamConnectionStatus.connected);
     _emit(
       StreamRecoveryResult<T>(
@@ -174,6 +195,9 @@ class StreamRecoveryWrapper<T> {
   void _onRealtimeError(Object error, [StackTrace? stack]) {
     if (_disposed) return;
 
+    // Cancel any stability timer — the connection is not stable.
+    _stabilityTimer?.cancel();
+
     final isRateLimit = _isRateLimitError(error);
     final isTimeout = _isTimeoutError(error);
     final tag = isRateLimit
@@ -337,6 +361,7 @@ class StreamRecoveryWrapper<T> {
     _disposed = true;
     _pollingTimer?.cancel();
     _recoveryTimer?.cancel();
+    _stabilityTimer?.cancel();
     _realtimeSub?.cancel();
     // Ensure the channel is removed from the recovering set when the
     // wrapper is torn down (e.g. provider disposed during navigation).