Predicting Properties of Periodic Systems from Cluster Data: A Case Study of Liquid Water