Finding the Homology of Decision Boundaries with Active Learning