Test file for the min cost flow algorithms.

Small fixes and doc improvements in MinMeanCycle.

Avoid map copy in MinCostMaxFlow.

Query improvements in the min cost flow algorithms.

- External flow and potential maps can be used.
- New query functions: flow() and potential().
- CycleCanceling also provides dual solution (node potentials).
- Doc improvements.

Bug fixes

Bug fixes in min cost flow files.
Use enum type instead of static constants in NetworkSimplex to avoid
linker errors.

Put the definitions of global functions into a .cc file.

Add a cost scaling min cost flow algorithm.

Add a cost scaling algorithm, which is performing generalized
push-relabel operations. It is almost as efficient as the capacity
scaling algorithm, but slower than network simplex.

Improvements in MinCostFlow and MinCostMaxFlow.

Main changes:
- MinCostMaxFlow also provides dual solution.
- Change the name of private members to start with "_".
- Change the name of function parameters not to start with "_".
- Remove unnecessary documentation for private members.
- Doc improvements.

Major improvements in NetworkSimplex.

Main changes:
- Use -potenital[] instead of potential[] to conform to the usual
terminology.
- Use function parameter instead of #define commands to select pivot rule.
- Use much faster implementation for the candidate list pivot rule.
It is about 5-20 times faster now.
- Add a new pivot rule called "Limited Search" that is a modified
version of "Block Search". It is about 25 percent faster on rather
sparse graphs.
- By default "Limited Search" is used for sparse graphs and
"Block Search" is used otherwise. This combined method is the most
efficient on every input class.
- Change the name of private members to start with "_".
- Change the name of function parameters not to start with "_".
- Remove unnecessary documentation for private members.
- Many doc improvements.