Numerical error cause invalid input to dijkstra algorithm in min cost max flow code

[Dmitrii Marin <dmitry.marin@…>]

Line 46 of successive_shortest_path_nonnegative_weights.hpp computes a new weight for Dijkstra algorithm. The weigh is guaranteed to be non-negative in theory. In practice numerical rounding errors may make it a small negative number causing the failure of Dijkstra algorithm.

Do something like this for floating point types: return std::max(value_type(0), get(distance_, source(v, g_)) - get(distance_, target(v, g_)) + get(weight_, v));