[SPARK-18412][SPARKR][ML] Fix exception for some SparkR ML algorithms training on libsvm data

## What changes were proposed in this pull request?
* Fix the following exceptions which throws when ```spark.randomForest```(classification), ```spark.gbt```(classification), ```spark.naiveBayes``` and ```spark.glm```(binomial family) were fitted on libsvm data.
```
java.lang.IllegalArgumentException: requirement failed: If label column already exists, forceIndexLabel can not be set with true.
```
See [SPARK-18412](https://issues.apache.org/jira/browse/SPARK-18412

) for more detail about how to reproduce this bug.
* Refactor out ```getFeaturesAndLabels``` to RWrapperUtils, since lots of ML algorithm wrappers use this function.
* Drop some unwanted columns when making prediction.

## How was this patch tested?
Add unit test.

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #15851 from yanboliang/spark-18412.

(cherry picked from commit 07be232e)
Signed-off-by: Yanbo Liang <ybliang8@gmail.com>