Variance Reduction in Bipartite Experiments through Correlation Clustering