Add durable pending runs and recipe list debugging

- Store pending runs in PostgreSQL for durability across restarts
- Add recovery method for orphaned runs
- Increase Celery result_expires to 7 days
- Add task_reject_on_worker_lost for automatic re-queuing
- Add logging to recipe list to debug filter issues

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

app/services/recipe_service.py

@@ -46,23 +46,30 @@ class RecipeService:
 
     async def list_recipes(self, actor_id: str = None, offset: int = 0, limit: int = 20) -> list:
         """
-        List available recipes.
+        List available recipes for a user.
+
+        L1 data is isolated per-user - only shows recipes owned by actor_id.
 
         Note: This scans the cache for recipe files. For production,
         you might want a database index of recipes by owner.
         """
+        import logging
+        logger = logging.getLogger(__name__)
+
         # Get all cached items and filter for recipes
-        # This is a simplified implementation - production would use a proper index
         recipes = []
 
         # Check if cache has a list method for recipes
         if hasattr(self.cache, 'list_by_type'):
             items = self.cache.list_by_type('recipe')
+            logger.info(f"Found {len(items)} recipe items in cache")
             for content_hash in items:
                 recipe = await self.get_recipe(content_hash)
                 if recipe:
-                    # Filter by actor if specified
-                    if actor_id is None or recipe.get("uploader") == actor_id:
+                    uploader = recipe.get("uploader")
+                    logger.info(f"Recipe {content_hash[:12]}: uploader={uploader}, actor_id={actor_id}")
+                    # Filter by actor - L1 is per-user
+                    if actor_id is None or uploader == actor_id:
                         recipes.append(recipe)
 
         # Sort by name