[SPARK-11349] [ML] Support transform string label for RFormula

[yanboliang]

Currently RFormula can only handle label with NumericType or BinaryType (cast it to DoubleType as the label of Linear Regression training), we should also support label of StringType which is needed for Logistic Regression (glm with family = "binomial").
For label of StringType, we should use StringIndexer to transform it to 0-based index.

[SparkQA]

Test build #44437 has finished for PR 9302 at commit 233ce5f.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[yanboliang]

I think further more we should provide a param named family for RFormula to indicate the estimator/model which will be applied to the dataframe transformed by this RFormula transformer, and then we can do more strict label validation check.

[mengxr]

This should not happen inside transform. we cannot afford one fit call per transform. We should inside the label indexer into the RFormula.fitinstead of RFormulaModel.transform.

[yanboliang]

Currently disable this assert due to StringIndexer output inconsistent nullable, I opened SPARK-11478 to track that issue. After it resolved, this assert will be enabled.

[SparkQA]

Test build #44925 has finished for PR 9302 at commit 40b9a27.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[mengxr]

LGTM. Merged into master. Thanks!