[SPARK-13676] Fix mismatched default values for regParam in LogisticRegression

## What changes were proposed in this pull request?

The default value of regularization parameter for `LogisticRegression` algorithm is different in Scala and Python. We should provide the same value.

**Scala**
```
scala> new org.apache.spark.ml.classification.LogisticRegression().getRegParam
res0: Double = 0.0
```

**Python**
```
>>> from pyspark.ml.classification import LogisticRegression
>>> LogisticRegression().getRegParam()
0.1
```

## How was this patch tested?
manual. Check the following in `pyspark`.
```
>>> from pyspark.ml.classification import LogisticRegression
>>> LogisticRegression().getRegParam()
0.0
```

Author: Dongjoon Hyun <dongjoon@apache.org>

Closes #11519 from dongjoon-hyun/SPARK-13676.