fix: Include node_states in SSE execution_complete and execution_failed events

- Serialize NodeExecutionState models to dict for JSON transmission
- Add node_states to execution_complete event for debug panel
- Add node_states to execution_failed event for partial progress visibility
- Enables WorkflowDebugPanel to display per-node debug information

Fixes: WorkflowDebugPanel not appearing due to missing node_states in SSE stream

backend/workflow_executor.py

@@ -216,6 +216,7 @@ async def execute_workflow(
                 "execution_id": execution_id,
                 "status": "completed",
                 "aggregated_result": aggregated,
+                "node_states": [ns.model_dump() for ns in node_states],  # Serialize Pydantic models
                 "total_nodes": len(node_states),
                 "completed_nodes": len([ns for ns in node_states if ns.status == NodeStatus.EXECUTED]),
                 "skipped_nodes": len([ns for ns in node_states if ns.status == NodeStatus.SKIPPED]),
@@ -238,12 +239,16 @@ async def execute_workflow(
 
         # NEW: Progress-Callback für Fehler
         if progress_callback:
-            await progress_callback("execution_failed", {
+            error_data = {
                 "execution_id": execution_id,
                 "status": "failed",
                 "error": str(e),
                 "completed_at": datetime.utcnow().isoformat()
-            })
+            }
+            # Include node_states if available (to show partial progress)
+            if 'node_states' in locals() and node_states:
+                error_data["node_states"] = [ns.model_dump() for ns in node_states]
+            await progress_callback("execution_failed", error_data)
 
         # Speichere Failed State
         completed_at = datetime.utcnow().isoformat()