[SPARK-38243][PYTHON][ML] Fix pyspark.ml.LogisticRegression.getThreshold error message logic

[zero323]

What changes were proposed in this pull request?

This PR replaces incorrect usage of str.join on a List[float] in LogisticRegression.getThreshold.

Why are the changes needed?

To avoid unexpected failure if method is used in case of multi-class classification.

After this change, the following code:

from pyspark.ml.classification import LogisticRegression

LogisticRegression(thresholds=[1.0, 2.0, 3.0]).getThreshold()

raises

Traceback (most recent call last):
  Input In [4] in <module>
    model.getThreshold()
  File /path/to/spark/python/pyspark/ml/classification.py:999 in getThreshold
    raise ValueError(
ValueError: Logistic Regression getThreshold only applies to binary classification, but thresholds has length != 2.  thresholds: [1.0, 2.0, 3.0]

instead of current

Traceback (most recent call last):
  Input In [7] in <module>
    model.getThreshold()
  File /path/to/spark/python/pyspark/ml/classification.py:1003 in getThreshold
    + ",".join(ts)
TypeError: sequence item 0: expected str instance, float found

Does this PR introduce any user-facing change?

No. Bugfix.

How was this patch tested?

Manual testing.

[srowen]

Could you do a quick search to see if .join(ts) is called like this anywhere else?

[zero323]

Thanks for the review @srowen.

Could you do a quick search to see if .join(ts) is called like this anywhere else?

I did a quick sweep, but nothing popped out.

In core, sql and most of ml, errors like this should be already detected with inline hints, ml.classification is just not there yet.

[srowen]

Merged to master

[zero323]

Thanks!