Given a markdown link like:

```md
[foo](foo bar.jpg)
```

Commonmark parsers will not recognise it as a link, because of the
space. To solve this, we have a Banzai filter that uses a regular
expression to detect the unparsed link and rewrite it into a parseable
one, replacing the space with '%20'. However, the use of backtracking
in the regular expression makes it vulnerable to complex inputs.

Switching to Gitlab::UntrustedRegexp lets us guarantee that the regular
expression will always complete in linear time, solving the resource
use issue.

Changelog: security