Concept inconsistency in dermoscopic concept bottleneck models: a rough-set analysis of the Derm7pt dataset