Agent can not stop after Kubeflow job is finished

With new training operator replaced, the Kubeflow CRD status conditions reason description has changed. We need to modify the processing of the state, so that the agent can stop correctly.