This paper addresses the limitations of existing federated recommendation systems (FedRecs) by proposing a user-governed data contribution architecture. Current FedRecs assume all users have the same 0-privacy budget, meaning they do not upload any data to the server, which overlooks users who are willing to share data for better recommendations. To bridge this gap, the paper introduces CDCGNNFed, a cloud-device collaborative graph neural network model. This model trains user-centric ego graphs locally and high-order graphs based on shared data collaboratively via contrastive learning. A graph mending strategy is also employed to predict missing links, enhancing the model's ability to capture high-order graph structural information. Extensive experiments on two public datasets demonstrate the effectiveness of the proposed method, showing superior performance compared to existing FedRecs and baseline methods. The main contributions include the introduction of a flexible and personalized privacy framework, the CDCGNNFed model, and the validation of its effectiveness through extensive experiments.This paper addresses the limitations of existing federated recommendation systems (FedRecs) by proposing a user-governed data contribution architecture. Current FedRecs assume all users have the same 0-privacy budget, meaning they do not upload any data to the server, which overlooks users who are willing to share data for better recommendations. To bridge this gap, the paper introduces CDCGNNFed, a cloud-device collaborative graph neural network model. This model trains user-centric ego graphs locally and high-order graphs based on shared data collaboratively via contrastive learning. A graph mending strategy is also employed to predict missing links, enhancing the model's ability to capture high-order graph structural information. Extensive experiments on two public datasets demonstrate the effectiveness of the proposed method, showing superior performance compared to existing FedRecs and baseline methods. The main contributions include the introduction of a flexible and personalized privacy framework, the CDCGNNFed model, and the validation of its effectiveness through extensive experiments.