Redundancy-aware unsupervised rankings for collections of gene sets