On the nature of the finite-temperature transition in QCD