Demojize performance

[cvzi]

This pull request improves the demojize performance. I have replace the big regular expression with a search tree. The method _get_search_tree() creates the tree once on first use. I have put an example of a tree in the comment of _get_search_tree()

The get_emoji_regexp() still exists but it is unused now.

Performance is faster on my machine for both single emoji and a long text. This is approximately how much faster it is on my machine:

Python	2.7	3.6	3.10
Single emoji:	7x	7x	12x	faster
Text [200k chars]:	1.5x	7x	23x	faster

---

I have also added emojize(language='alias') as an alternative to emojize(use_aliases=True).
When someone uses emojize(language='de', use_aliases=True) or any other language then 'en', a warning is shown:

warnings.warn("use_aliases=True is only supported for language='en'. "
              "It is recommended to use emojize(string, language='alias') instead")

(I could remove this change to the aliases part, it is not connected to the performance at all)

[TahirJalilov]

Thank you for your PR @cvzi