Pareto fronts for least interesting solution

[matthieu-vergne]

For now, if the max limit is reached, we remove the worst solution based solely on the value assigned to the solution, without considering its optimality. I think that using a Pareto front to find the best and the most recent solutions can be more reliable:

1. S = {solutions}
2. F = ParetoFront(S)
3. S = S \ F
4. restart from 2 until S is empty
5. select a random candidate in F to remove

Or something similar.