refactor(api): move llm quota deduction to app graph layer (#32786)

api/core/workflow/nodes/loop/loop_node.py

@@ -413,6 +413,7 @@ class LoopNode(LLMUsageTrackingMixin, Node[LoopNodeData]):
 
     def _create_graph_engine(self, start_at: datetime, root_node_id: str):
         # Import dependencies
+        from core.app.workflow.layers.llm_quota import LLMQuotaLayer
         from core.app.workflow.node_factory import DifyNodeFactory
         from core.workflow.entities import GraphInitParams
         from core.workflow.graph import Graph
@@ -454,5 +455,6 @@ class LoopNode(LLMUsageTrackingMixin, Node[LoopNodeData]):
             command_channel=InMemoryChannel(),  # Use InMemoryChannel for sub-graphs
             config=GraphEngineConfig(),
         )
+        graph_engine.layer(LLMQuotaLayer())
 
         return graph_engine