It seems libicu-dev no longer installs libicu55 needed by charlock_holmes.
Fix builds that fail on Heroku.
* Specs for language detection
* Use CharlockHolmes instead of NKF
* Correct mistakes
* Correct style
* Set hint_enc instead of falling back and strip_tags
* Improve specs
* Add dependencies
This change adds mp4 support confirmed on Heroku-16 stack.
Use https://github.com/heroku/heroku-buildpack-apt to
install protobuf-compiler and libprotobuf-dev which are needed by cid3