Stackoverflow NER dataset cleaning

[symeneses]

It adds entity mappings, removes sentences with metadata and adds questions/answers counts.
It fixes #2228.

Expected Results:

from flair.data import Corpus
from flair.datasets import STACKOVERFLOW_NER

corpus: Corpus = STACKOVERFLOW_NER()
print(corpus)

Corpus: 9263 train + 2896 dev + 3108 test sentences

corpus.train[0:3]

[Sentence: "If I would have 2 tables" [− Tokens: 6 − Token-Labels: "If I would have 2 tables <S-Data_Structure>"],
Sentence: "How do I get this result" [− Tokens: 6],
Sentence: "The following query needs to be adjusted , but I dont know how" [− Tokens: 13]]

Notes:
It would be nicer to extend the parameter comment_symbol in the ColumnCorpus class or create one similar that accepts also alist. In this case, the new parameter would take the values in the list disallowed_list.

[alanakbik]

@symeneses thanks a lot for spotting and fixing this!

Good idea with extending the comment_symbol into a disallowed_list! Care to do a PR for this? ;)

[symeneses]

I could do that and then use it to clean this particular dataset. 👍🏽