Does finite-temperature decoding deliver better optima for noisy Hamiltonians?