Folded and log-folded-t distributions as models for insurance loss data