Greedier solution recreation process

[jcoupey]

The try_job_addition worst-case complexity is $O(U^2\times R)$ where $U$ is the number of unassigned jobs we try to re-insert, and $R$ is the number of candidate routes for re-insertion. This is not a big deal in typical re-insertion tries after applying a local search operator because then $R\le 2$ and $U$ is usually small.

On the other hand, when we ruin a bigger part of the solution and want a full recreation, then $R$ is the number of existing vehicles in the current solution and $U$ can be much higher. So the time spent doing the recreation can start to be significant, especially in the light of #874 where we'll probably tend to remove more jobs than we currently do.

We could have a pre-established insertion order for unassigned jobs, based say on priority, capacity, TW narrowness, closeness to some vehicle or similar. In which case the above complexity would drop to $O(U\times R)$.