toolbarl: long long costs

- taking into account initial lower bound in problem wcsp file if any
- do not allocate memory for depth-first branch and bound stuff if using BE only (larger problems can be solved)

These are a number of WCSPs derived from two model-based diagnosis and planning problems. The first problem is, given a plan currently in execution and partial observations, to compute the plan's success probability (based on a probabilistic system model, hence the need for soft constraints). The second problem is to correctly diagnose a fault and robustly plan around this contingency within a hybrid system, i.e. a system modeled continuously (differential eq.) and discrete (stochastic automaton). The solution is to discretize the hybrid model, generating an abstract discrete model which can than be turned into a WCSP. See the problem description for a detailed account of the scenarios.

modified the if test for the BEMode to include DomainAbstractionBEMode; now the TreeDecomposition is accessable for DomainAbstraction, too

 * Tree Decomposition can now be used in conjunction with BEDA
 * before search, the lower bound found by BEDA is displayed

 + the actual code
 + external code (heap, CuTest)
 + helper scripts
 + license files

Changes in this file are to allow choosing the BEDA algorithm through the command line option -j. The additional option -M is used to choose a size limit for the messages computed during BEDA inference.

toolbarTest.py is a simple python script running some tests on the BEDA algorithm implemented in toolbar. result.py contains some helper functions and classes used by toolbarTest.py.

MPE instances generated with 0.05 error prior, 10e-3 precision, and allele frequency from data.

le petit est resolu jusq'a l'optimum par vac!

     toulbar2 rectanglecoverS.wcsp 500000 C10000A10000

      Read 208 variables, with 10 values at most, and 595 constraints.
      Lb before VAC: 60000
      Preprocessing VAC mean lb/incr: 763.206     total increments: 131
      Lb after VAC: 159980

   et l'optimum c'est efectivement 16