Allow word tokenization override

[freecraver]

See #26 (comment)

Introduces a non-breaking change which allows to override custom word-level tokenization.

The new f_tokenize_words argument accepts a function which maps a text to its words.

example:

from nltk import word_tokenize
r = Readability(text, f_tokenize_words=word_tokenize)

Tests run ✔️
Tests added ✔️
Added section 'What makes a word' to Readme ✔️

Additional remarks:

• The main difference between nltks TweetTokenizer and the TreebankWordTokenizer I observed is the handling of clitics and abbreviations:

Text	Tweet	Treebank
"We've got two different solutions"	["We've", 'got', 'two', 'different', 'solutions']	['We', "'ve", 'got', 'two', 'different', 'solutions']
'How common are abbreviations in the U.S.?'	['How', 'common', 'are', 'abbreviations', 'in', 'the', 'U', '.', 'S', '.', '?']	['How', 'common', 'are', 'abbreviations', 'in', 'the', 'U.S.', '?']