Configurable exploration level

[jcoupey]

The CVRP implementation leaves some open choices regarding the trade-off between quality and computing times. Main factors currently at stake are:

• number of parameters used for the clustering heuristic (related Tune CVRP heuristic parameters #120)
• number of jobs popped from routes to get out of local minimum during local search (see Local-search phase for CVRP #92 (comment))

Given that the trade-off choice is highly dependent on use-case and/or instance size, it's hard to have an all-round choice here. Instead, it would be interesting to provide a high-level tuning, say a kind of exploration level that could be defined in a 0..n range. Increasing that level would mean longer computing times in the hope of getting better solutions.

[jcoupey]

Done in #132.