Top-k overlapping densest subgraphs: approximation algorithms and computational complexity