Tests

We had problems with the tests in t/Graph.t in the depth_first_search_levels because the returned data structure was an array ref based on values provided by **keys**. With hash order randomization in perl 5.18+, it threw the original tests into disarray and therefore new ones were written based on what a depth first search on the below graph would look like if the branching decisions were random.

      0
    /   \
   1     4
  / \   / \
 2   3 5   6
        \
         7