[SPARK-16710][SPARKR][ML] spark.glm should support weightCol

## What changes were proposed in this pull request?
Training GLMs on weighted dataset is very important use cases, but it is not supported by SparkR currently. Users can pass argument ```weights``` to specify the weights vector in native R. For ```spark.glm```, we can pass in the ```weightCol``` which is consistent with MLlib.

## How was this patch tested?
Unit test.

Author: Yanbo Liang <ybliang8@gmail.com>

Closes #14346 from yanboliang/spark-16710.