[ML][PYTHON][SPARK-13008] Put one alg per line in pyspark.ml all lists

[jkbradley]

This is to fix a long-time annoyance: Whenever we add a new algorithm to pyspark.ml, we have to add it to the __all__ list at the top. Since we keep it alphabetized, it often creates a lot more changes than needed. It is also easy to add the Estimator and forget the Model. I'm going to switch it to have one algorithm per line.

This also alphabetizes a few out-of-place classes in pyspark.ml.feature. No changes have been made to the moved classes.

CC: @thunterdb

[SparkQA]

Test build #50121 has finished for PR 10927 at commit bb0f1ef.

• This patch fails Python style tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • class ChiSqSelector(JavaEstimator, HasFeaturesCol, HasOutputCol, HasLabelCol):
  • class ChiSqSelectorModel(JavaModel):
  • class PCA(JavaEstimator, HasInputCol, HasOutputCol):
  • class PCAModel(JavaModel):
  • class RFormula(JavaEstimator, HasFeaturesCol, HasLabelCol):
  • class RFormulaModel(JavaModel):

[SparkQA]

Test build #50123 has finished for PR 10927 at commit 639a562.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[mengxr]

LGTM but it has merge conflicts with master now.

[SparkQA]

Test build #52270 has finished for PR 10927 at commit 12b15fb.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[mengxr]

Merged into master. Thanks!