Support negative costs and bounds in NetworkSimplex (#270)

* The interface is reworked to support negative costs and bounds.
- ProblemType and problemType() are renamed to
SupplyType and supplyType(), see also #234.
- ProblemType type is introduced similarly to the LP interface.
- 'bool run()' is replaced by 'ProblemType run()' to handle
unbounded problem instances, as well.
- Add INF public member constant similarly to the LP interface.
* Remove capacityMap() and boundMaps(), see also #266.
* Update the problem definition in the MCF module.
* Remove the usage of Circulation (and adaptors) for checking feasibility.
Check feasibility by examining the artifical arcs instead (after solving
the problem).
* Additional check for unbounded negative cycles found during the
algorithm (it is possible now, since negative costs are allowed).
* Fix in the constructor (the value types needn't be integer any more),
see also #254.
* Improve and extend the doc.
* Rework the test file and add test cases for negative costs and bounds.

 #! /usr/bin/env python

 import sys

 from mercurial import ui, hg

 from mercurial import util

 util.rcpath = lambda : []

 if len(sys.argv)>1 and sys.argv[1] in ["-h","--help"]:

     print """

 This utility just prints the length of the longest path

 in the revision graph from revison 0 to the current one.

 """

     exit(0)

 u = ui.ui()

 r = hg.repository(u, ".")

 N = r.changectx(".").rev()

 lengths=[0]*(N+1)

 for i in range(N+1):

     p=r.changectx(i).parents()

     if p[0]:

         p0=lengths[p[0].rev()]

     else:

         p0=-1

     if len(p)>1 and p[1]:

         p1=lengths[p[1].rev()]

     else:

         p1=-1

     lengths[i]=max(p0,p1)+1

 print lengths[N]