LogisticRegression._checkThresholdConsistency should use values not Params

_checkThresholdConsistency incorrectly uses getParam in attempt to access threshold and thresholds values. Furthermore it calls it with Param instead of str:

>>> from pyspark.ml.classification import LogisticRegression
>>> lr = LogisticRegression(threshold=0.25, thresholds=[0.75, 0.25])
Traceback (most recent call last):
...
TypeError: getattr(): attribute name must be string

Finally exception message uses join without converting values to str.