* Add failing specs for hashtag and username extraction in language detector

* Remove usernames and hashtags from text before language detection

* Handle multiple instances of special case, and reduce whitespace