Bellman-Ford only detects presence of negative weight cycle,
not what that negative cycle is.

There's almost always a negative weight cycle, and in practice this
algorithm is useless, as indicated by the author of the blog post that 
inspired this code.

This is a solution to problem presented in the textbook "Introduction to 
Algorithms", by Cormen, Leiserson, Rivest, Stein,
in its chapter on the single-source shortest paths problem. To explain the
problem in terms of the Bellman-Ford algorithm, it is to:

- detect the existence of a negative weight cycle in the currency graph
- produce the path of any (not all) negative weight cycle in the graph

Negative weight cycle is determined by doing an extra relaxation step
after |V| - 1 relaxation steps.

After |V| - 1 relaxations, all paths from the source to each vertex are guaranteed
to be fixed, if there is no negative weight cycle. However, supposing
after |V| relaxations, the distance value for a vertex v has decreased,
then a negative weight cycle starts and ends at v.

Is there anything wrong with this logic?