[graph-tool] efficient way to load a numpy adjecency matrix?