perf: reduce network communication overhead cost on reduce step for LightGBM learners

[imatiach-msft]

Reduce the network communication overhead cost in the reduce step for LightGBM learners (LightGBMClassifier, LightGBMRegressor and LightGBMRanker) by only returning the model from one of the nodes (determined by order of nodes initialized). This is a small optimization that should slightly reduce the overall training time as well as make the code more robust by reducing overall network communication during reduce step.

[imatiach-msft]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[codecov]

Codecov Report

Merging #869 into master will increase coverage by 2.91%.
The diff coverage is 89.47%.

@@            Coverage Diff             @@
##           master     #869      +/-   ##
==========================================
+ Coverage   82.35%   85.26%   +2.91%     
==========================================
  Files         189      189              
  Lines        8693     8708      +15     
  Branches      517      518       +1     
==========================================
+ Hits         7159     7425     +266     
+ Misses       1534     1283     -251     

Impacted Files	Coverage Δ
...com/microsoft/ml/spark/lightgbm/LightGBMBase.scala	90.62% <ø> (ø)
...a/com/microsoft/ml/spark/lightgbm/TrainUtils.scala	87.36% <89.47%> (-0.04%)	⬇️
...icrosoft/ml/spark/downloader/ModelDownloader.scala	85.88% <0.00%> (-1.18%)	⬇️
.../microsoft/ml/spark/vw/VowpalWabbitRegressor.scala	70.00% <0.00%> (+10.00%)	⬆️
...a/com/microsoft/ml/spark/vw/VowpalWabbitBase.scala	89.04% <0.00%> (+15.06%)	⬆️
...ala/com/microsoft/ml/spark/image/UnrollImage.scala	77.65% <0.00%> (+17.02%)	⬆️
...icrosoft/ml/spark/lime/SuperpixelTransformer.scala	93.33% <0.00%> (+20.00%)	⬆️
...a/com/microsoft/ml/spark/vw/HasSumcollisions.scala	100.00% <0.00%> (+25.00%)	⬆️
...rosoft/ml/spark/core/schema/BinaryFileSchema.scala	100.00% <0.00%> (+25.00%)	⬆️
.../scala/com/microsoft/ml/spark/vw/VectorUtils.scala	96.42% <0.00%> (+28.57%)	⬆️
... and 13 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 840781a...a081ca5. Read the comment docs.