Online learning with graph-structured feedback against adaptive adversaries