llama model cache, a few bugfixes

backend/src/chat.py

@@ -91,7 +91,8 @@ class ChatView:
             filters["id__lte"] = self._earliest_message_id
 
         # Tortoise query: filter, order, and limit
-        records_rev = await db.Message.filter(**filters).order_by("-time").limit(load_count)
+        # ordering by id is more robust than by time
+        records_rev = await db.Message.filter(**filters).order_by("-id").limit(load_count)
 
         # Update the earliest id for the next load operation
         if records_rev: