Regret optimality in semi-Markov decision processes with an absorbing set