* Implement ```MLWriter/MLWritable/MLReader/MLReadable``` for PySpark.
* Making ```LinearRegression``` to support ```save/load``` as example. After this merged, the work for other transformers/estimators will be easy, then we can list and distribute the tasks to the community.

cc mengxr jkbradley

Author: Yanbo Liang <ybliang8@gmail.com>
Author: Joseph K. Bradley <joseph@databricks.com>

Closes #10469 from yanboliang/spark-11939.