a few quality of life improvements

[Jivanni]

Hi metalcorebear, I have made some changes to your lovely library after using it in some of my projects:

• the NRCLex object now does not take the input text when initialized.
• I have added two methods, one to load the text as string (same behaviour as before) and one to load the text as a list of tokens. I think this is very useful when working with "dirty" text that needs more then tokenization and needs to be pre-processed beforehand.
• NRC dictionary is now loaded from a json file and object instantiation time (now you only need to instantiate once), the dictionary to use can be changed through the NRCLex constructor. We could add more languages in the future as json files.
• Rather then simply tokenizing with TextBlob, now the library takes advatage of the token.lemmatize() function to lemmatize each token. This improves the recall of the library as now it finds more words in the dictionary, therefore assigning better scores. This is at the cost of longer execution time.
• Added some docstrings.

If you this those changes make sense, please feel free to merge them into your project.

jivanni

[riyadhctg]

I actually think changes in this PR breaks the package installation (i.e. pip install git+......)

• setup.py doesn't include the json file so the package behavior is broken

Also, I think we don't need to install collections and json as separate package, at least for newer python version

[metalcorebear]

You could be right. The code is pretty messy and not well documented (I didn't expect it to become a popular project). I honestly haven't looked at the repo in the last year, but I know it needs some improvements. I'm hoping to find time in August to fix it up, but if you want to make the necessary improvements, please feel free to do so.