Some thoughts on declustering in probabilistic seismic-hazard analysis