Rank-into-rank hypotheses and the failure of GCH