Add support for Multi-person skintones

[cvzi]

Finally finished this. It looks like quite a lot of changes, but apart from the multi-person skintones, I mostly just moved code around and deleted unused code:

The logic from demojize() is moved to two separate private function tokenize() and filter_tokens() in a new file emoji/tokenizer.py

demojize(), replace_emoji(), emoji_list() and analyze() all use the same logic, that's why I moved it into the new private function tokenize() and filter_tokens(). Otherwise there would have been duplicated code. Also the logic for the search tree is moved to the emoji/tokenizer.py file.

A new public function analyze() is available and that supports the multi-person skintones. It's similar to emoji_list(). It returns Token and EmojiMatch objects that provide more information than emoji_list() and also allow to split up ZWJ-emoji into the sub-emoji.

The handling of the multi-person skintones in demojize() and replace_emoji can be controlled by the new emoji.config class, which is a static class that works as a module-wide configuration:

emoji.config.demojize_keep_zwj = True
emoji.config.replace_emoji_keep_zwj = True

---

I have removed support for Python 2, 3.4, 3.5, because I used some features of Python 3.6. I removed lots of things that were specific to Python 2, especially all the u'strings' .

Fixes #204 Fixes #256

[TahirJalilov]

Thank you, @cvzi for your job!

[lsmith77]

indeed thank you!