Use cost bound in heuristic

[jcoupey]

Issue

Fixes #1219

Tasks

• Store and update relevant data to compute the bound
• Avoid going through whole route for rank-based insertion tests based on bound
• Same but for pickup and delivery
• Try ordering unassigned jobs to increase bound efficiency
• Update CHANGELOG.md
• review

[jcoupey]

My attempts at ordering unassigned jobs to favor cutting branches were void: we'd need to maintain an extra data structure + pay the re-ordering price from time to time, which in my tests destroys the benefit of using the bound with unsorted data.

[jcoupey]

Benchmarking setup for job-based instances

Instances

In order to measure the impact of this PR, I've generated random instances with sizes ranging from 200 to 5000 jobs. Since route length is a major factor in complexity, I've generated several instances per number of jobs to match typical sizes: S (10 jobs per route), M (50 jobs per route), L (100 jobs per route), XL (250 jobs per route) and XXL (500 jobs per route).

Process

For all those instances I'm solving only with the heuristics (no local search) with both current master at 9905815 and PR at 31476a7. We're interested in the solving time ratio.

Results

Solving speedup between master and PR depending on instance and route sizes.

Jobs	S	M	L	XL	XXL
200	1.50	1.71	1.58
500	1.43	2.09	1.87	1.64
1000	1.86	1.73	1.82	1.71	1.69
2000	1.68	2.65	2.36	2.31	1.86
3000	1.92	2.99	2.61	2.68	1.84
4000	1.93	3.02	2.53	2.59	1.88
5000	2.04	3.25	3.04	2.48	2.02

If we look at total computing time, the speedup is 2.17, meaning PR is more than 2x faster than master. Again that is only on heuristic time, which may be a small part of overall computing time is many situations.

[jcoupey]

Benchmarking setup for shipment-based instances

Using the usual Li&Lim benchmark (all classes with instance sizes from 100 to 1000), again solving only with the heuristics (no local search) with both current master at 9905815 and PR at 31476a7.

Average computing times (ms) and speedup

	9905815	31476a7	Speedup
100	5	4	1.25
200	25	21	1.19
400	109	85	1.28
600	291	201	1.45
800	599	372	1.61
1000	984	581	1.69