fix: notification retry queue - resolve hour-long delays

- Pending notifications now send immediately (no backoff)
- Backoff uses created_at instead of updated_at (fixed timeline)
- Increase batch size from 50 to 200 for faster queue drain
- Add queue depth logging and skip debug logging

worker/src/main.py

@@ -59,20 +59,33 @@ async def process_notification_queue(pool: asyncpg.Pool, bot: ExtBot) -> int:
     """Process pending notifications from the retry queue."""
     rows = await pool.fetch("""
         SELECT id, ad_id, telegram_id, message_text, type, attempts, 
-               max_attempts, last_error, updated_at
+               max_attempts, last_error, updated_at, created_at
         FROM notification_queue
         WHERE status IN ('pending', 'failed')
-        ORDER BY attempts ASC, updated_at ASC
+        ORDER BY attempts ASC, created_at ASC
-        LIMIT 50
+        LIMIT 200
     """)
 
+    # Log queue depth
+    queue_stats = await pool.fetchrow("SELECT count(*) FROM notification_queue WHERE status IN ('pending', 'failed')")
+    total_queued = queue_stats["count"]
+    logger.info("Notification queue: %d total, processing up to %d", total_queued, len(rows))
+
     processed = 0
 
     for row in rows:
-        backoff_min = min(2 ** row["attempts"], 60)
+        # Pending (never tried) — send immediately
-        retry_after = row["updated_at"] + timedelta(minutes=backoff_min)
+        if row["status"] == "pending":
+            backoff_min = 0
+        else:
+            # Failed — exponential backoff from created_at (fixed timeline)
+            backoff_min = min(2 ** row["attempts"], 60)
+
+        retry_after = row["created_at"] + timedelta(minutes=backoff_min)
 
         if datetime.now(tz=timezone.utc) < retry_after:
+            logger.debug("Skipping notification %s — retry after %s (now %s)", 
+                         row["id"], retry_after, datetime.now(tz=timezone.utc))
             continue
 
         try: