Instead of rendering this value server-side, we use Javascript and Gon
to apply the user's color scheme (or the default) to any syntax
highlighted code blocks.

This will make it easier to cache these blocks in the future because
they're no longer state-dependent.

It was redundant because `gfm` also took options.

I have mainly created the rugments fork for the purpose of improving
gitlab's highlighting. Nowadays IMO it works way better than the old
highlight.js solution. But the development is stuck on my side because
of a couple of personal reasons:

  * I have finished my studies; last months I was writing my master
    thesis. So there was a huge time problem. I am sorry for that.

  * I had to move to Munich due to getting a (paid) job. Searching a
    flat here is horrible... :)

  * Last but not least, maintaining the same code base in two seperate
    projects is a mess.

I have decided to switch back to rouge due to several reasons:

  * In the beginning I was quite motivated, but since I start
    working on my new job next week, the best solution IMO is
    switching back to upstream rouge.

  * Rouge is continously improving:
    https://github.com/jneen/rouge/blob/master/CHANGELOG.md
    http://rouge.jneen.net/

  * There should be absolutely no regressions with this change. Most
    likely this pull request will almost fix some minor bugs.

  * One less gem in gitlab is a good thing. since Gitlab is quite a
    huge bundle of gems. Reducing complexity should be a major
    milestone.

Thanks a lot to @stanhu and @jneen for the review!

While `escape_once` will create this HTML entity, Nokogiri always
translates it back before it gets to IssueReferenceFilter, so there
should be no danger of erroneous issue links.