Implement MNMG Bellman-Ford algorithm to compute shortest paths in the presence of 0 and negative weights

[ChuckHastings]

Our current shortest path implementation uses delta stepping to parallelize Dijkstra's algorithm efficiently on the GPU.

Dijkstra's algorithm requires all edges to have positive weights. In order to support weight 0 edges and negative weight edges we need to implement the Bellman-Ford algorithm.

This issue is to provide a C++ implementation of Bellman-Ford.

[eriknw]

Dijkstra's algorithm requires all edges to have positive weights. In order to support weight 0 edges and negative weight edges we need to implement the Bellman-Ford algorithm.

The GraphBLAS implementation of SSSP in LAGraph uses delta stepping, and it supports negative weights. See for instance: https://github.com/GraphBLAS/LAGraph/blob/afeb9dffca463fb966f1d9637986af55b04df461/src/algorithm/LAGr_SingleSourceShortestPath.c#L364-L376