[MRG] EHN add voting paramter for ClusterCentroids

[glemaitre]

Reference Issue

closes #317

#316 needs to be merged before and this PR needs to be rebased.

What does this implement/fix? Explain your changes.

Allow to use nearest neighbors instead of centroids themselves.
More efficient for sparse input.

Any other comments?

[glemaitre]

@massich @chkoar As discussed in #316, here it comes and ready for revision.

[glemaitre]

Merge #316 before review, it will reduce the number of changes.

[codecov]

Codecov Report

Merging #318 into master will increase coverage by 0.11%.
The diff coverage is 97.4%.

@@            Coverage Diff             @@
##           master     #318      +/-   ##
==========================================
+ Coverage   98.14%   98.25%   +0.11%     
==========================================
  Files          66       66              
  Lines        4043     3956      -87     
==========================================
- Hits         3968     3887      -81     
+ Misses         75       69       -6

Impacted Files	Coverage Δ
imblearn/ensemble/easy_ensemble.py	100% <ø> (ø)	⬆️
imblearn/over_sampling/base.py	100% <100%> (ø)	⬆️
imblearn/over_sampling/random_over_sampler.py	100% <100%> (ø)	⬆️
...ototype_generation/tests/test_cluster_centroids.py	100% <100%> (ø)	⬆️
...arn/under_sampling/prototype_selection/nearmiss.py	98.66% <100%> (ø)	⬆️
imblearn/ensemble/balance_cascade.py	100% <100%> (ø)	⬆️
...sampling/prototype_generation/cluster_centroids.py	100% <100%> (ø)	⬆️
...prototype_selection/instance_hardness_threshold.py	97.33% <100%> (ø)	⬆️
...g/prototype_selection/edited_nearest_neighbours.py	100% <100%> (ø)	⬆️
.../under_sampling/prototype_selection/tomek_links.py	100% <100%> (ø)	⬆️
... and 51 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cddf39b...6c34e56. Read the comment docs.

[glemaitre]

@massich I am going to rebase once the #316 will be merged.