[SPARK-11987] Python API update for ChiSqSelector and QuantileDiscretizer

[yinxusen]

https://issues.apache.org/jira/browse/SPARK-11987

For QuantileDiscretizer, I add a QuantileDiscretizerModel in it as a nested class. The reason is a QuantileDiscretizer generates a Bucketizer mode, while the latter extends from a JavaTransformer in PySpark rather than a JavaModel, which prevents us transfering java model to python model easily. It also prevents us to call _call_java("get_splits") from outside since it is a private method.

I tried to rewrite the Bucketizer into a JavaModel in the first time, but it may hurt the current interfaces since JavaModel and JavaTransformer have different ways to set uid and different construction method args __init__(...)

[SparkQA]

Test build #46772 has finished for PR 10007 at commit 05f3edd.

• This patch fails PySpark unit tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):\n * class QuantileDiscretizer(JavaEstimator, HasInputCol, HasOutputCol):\n * class QuantileDiscretizerModel(JavaModel):\n * class ChiSqSelector(JavaEstimator, HasFeaturesCol, HasOutputCol, HasLabelCol):\n * class ChiSqSelectorModel(JavaModel):\n

[SparkQA]

Test build #46802 has finished for PR 10007 at commit 3789867.

• This patch fails PySpark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #47136 has finished for PR 10007 at commit 3a33327.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):\n * class QuantileDiscretizer(JavaEstimator, HasInputCol, HasOutputCol):\n * class QuantileDiscretizerModel(JavaModel):\n * class ChiSqSelector(JavaEstimator, HasFeaturesCol, HasOutputCol, HasLabelCol):\n * class ChiSqSelectorModel(JavaModel):\n

[yinxusen]

close it for duplication