SPARKNLP 643 detecting spark version in a safer way

[maziyarpanahi]

Some environments will have an alphanumeric version of Spark. This PR will make a safer comparison when it comes to whether or not we are in Spark 3.3.0 environment (needed for some operations).

We could use

cleaned_version = re.findall(r'(?:(\d+\.(?:\d+\.)*\d+))', self.sc.version)

or LooseVersion from distutils.version

LooseVersion(self.spark_version()) >= LooseVersion("3.3.0")

Tested on EMR:

>>> from sparknlp.training import CoNLL
>>> trainingData = CoNLL().readDataset(spark, "file:///home/hadoop/en.test.lemma.conllu")
>>>

[maziyarpanahi]

@danilojsl I have 2 commits representing both approaches, they both work even in EMR. Please let me know which one you are more comfortable with

[danilojsl]

@maziyarpanahi the one with LooseVersion LGFM