Function: Use BitSet for used_vars about ibex-lib

[gchabert]

Unfortunately this has introduced a loss of efficiency.

Between commit da73630 and commit 6feec93, we see more than 10% loss. This has been tested on Bertrand's optimizer04 as follows:

./optimizer04 ../plugins/optim/benchs/coconutbenchmark-library2/hs099.nl acidhc4 compo lsmearmg bs 1 1.e-7 1.e-6 7200 1

with the initial domain restricted as follows

double _sol[]={0.541993 , 0.530903 , 0.508103 , 0.480116 , 0.451103 , 0.409231 , 0.35281 , -1939.93 , -7908.68 , -18412.6 , -20003.9 , -17422.7 , 
20342.1 , -155.195 , -322.304 , -97.854 , 34.2024 , 69.0438 , 770.176 };
Vector sol(19,_sol);
sys->box = sol + Interval(-1,1)*abs(sol);

Output is:

 best bound in: [-831080315.4147157,-831079484.3352315]
 relative precision obtained on objective function: 9.999999998665447e-07  [passed]
 absolute precision obtained on objective function: 831.0794842243195  [failed]
 best feasible point: (0.5412766277476286 ; 0.5294559925796978 ; 0.5088739226806065 ; 0.4805469809611225 ; 0.4499308972695276 ; 0.4097497484257161 ; 0.3525770351404032 ; -1949.522701654927 ; -7956.944777032146 ; -18514.08979270156 ; -20059.50647496983 ; -17495.46201687364 ; 20253.57237668331 ; -155.9618161241145 ; -324.6319498977834 ; -97.65385072035226 ; 35.83718343826147 ; 66.72459489422636 ; 772.1428360825062)
 number of cells: 2054

Time with da73630 is around 21.3s.
Time with 6feec93 is around 19.3s.

The new version however makes use of vector instead of raw arrays.