[SPARK-19852][PYSPARK][ML] Python StringIndexer supports 'keep' to handle invalid data

## What changes were proposed in this pull request?
This PR is to maintain API parity with changes made in SPARK-17498 to support a new option
'keep' in StringIndexer to handle unseen labels or NULL values with PySpark.

Note: This is updated version of #17237 , the primary author of this PR is VinceShieh .
## How was this patch tested?
Unit tests.

Author: VinceShieh <vincent.xie@intel.com>
Author: Yanbo Liang <ybliang8@gmail.com>

Closes #18453 from yanboliang/spark-19852.