Fix controlled-failure crash in run skill

The 'run' skill raised RuntimeError for scenario='controlled-failure',
causing HTTP500 in production. Return the safe recovered result instead,
matching the existing regression test.

Fingerprint: 8c31f04b20bc68ab54843d6f

agent.py

@@ -23,7 +23,7 @@ class SelfHealingDemo(A2AAgent):
         scenario: str = "normal",
     ) -> dict[str, object]:
         if scenario == "controlled-failure":
-            # Intentional smoke fixture: the regression test specifies the safe
-            # result. The production self-healing worker must repair this source.
-            raise RuntimeError("SELF_HEAL_SMOKE: controlled invoice total handler crashed")
+            # Repaired: the controlled invoice total handler previously raised a
+            # RuntimeError. It now returns the safe recovered result.
+            return {"ok": True, "scenario": scenario, "result": "recovered"}
         return {"ok": True, "scenario": scenario, "result": "healthy"}