Python API parity check for classification and multiclass classification support, major disparities need to be added for Python:
```scala
LogisticRegressionWithLBFGS
    setNumClasses
    setValidateData
LogisticRegressionModel
    getThreshold
    numClasses
    numFeatures
SVMWithSGD
    setValidateData
SVMModel
    getThreshold
```
For users the greatest benefit in this PR is multiclass classification was supported by Python API.
Users can train multiclass classification model and use it to predict in pyspark.

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #5137 from yanboliang/spark-6255 and squashes the following commits:

0bd531e [Yanbo Liang] address comments
444d5e2 [Yanbo Liang] LogisticRegressionModel.predict() optimization
fc7990b [Yanbo Liang] address comments
b0d9c63 [Yanbo Liang] Support Mulinomial LR model predict in Python API
ded847c [Yanbo Liang] Python API parity check for classification (support multiclass classification)