[SPARK-10194] [MLlib] [PySpark] SGD algorithms need convergenceTol parameter in Python

[yanboliang]

SPARK-3382 added a convergenceTol parameter for GradientDescent-based methods in Scala. We need that parameter in Python; otherwise, Python users will not be able to adjust that behavior (or even reproduce behavior from previous releases since the default changed).

[SparkQA]

Test build #41619 has finished for PR 8457 at commit 92d6d8c.

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

[feynmanliang]

Not your fault, but if you make any additional changes can you add a "." at the end of this sentence?

[feynmanliang]

Also, the default values should be documented.

[feynmanliang]

LGTM overall

The formal parameter changes to PythonMLLibAPI will break binary compatibility, but this may not be a big issue since the class is private. Perhaps others will have better insight.

[mengxr]

chop down args

[yanboliang]

I open SPARK-10560 to track the StreamingLogisticRegressionWithSGD issues included what @feynmanliang mentioned in this PR.
Let's this PR focus on the topic. @mengxr

[SparkQA]

Test build #42317 has finished for PR 8457 at commit c2f3112.

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

[mengxr]

LGTM. Merged into master. Thanks!