Rethinking Iterative Proportional Fitting: Scalable And Hybrid Approaches To Joint Distribution Fitting