Add custom protocol whitelisting to SanitizationFilter

Addresses internal https://dev.gitlab.org/gitlab/gitlabhq/issues/2613

We allow any protocol for autolinks: irc://irc.freenode.net/git

But manual Markdown links with the same protocol get sanitized: `[This will not be clickable](irc://irc.freenode.net/git)`: [This will not be clickable](irc://irc.freenode.net/git)

To get around this we have to first allow *all* protocols, and then manually clean dangerous (i.e., `javascript:`) protocols.

See merge request !1496