The refactoring of e-mails in Review Board 3.0 ended up causing a

regression where URLs in e-mails would have the site root applied twice.

For example, for a server hosted at http://example.com/rb, the URL for

a review request would be generated as http://example.com/rb/rb/r/1.

This was due to no longer computing the site URL (which was the scheme +

domain only), but instead using the result of get_server_url() which

includes the site root. This regression wasn't apparent due to a lack of

testing on configurations where the site root was anything other than

/.
Now, we pass in the correct value for site_url, computing it from

build_server_url for the / URL (which clobbers the site root) and

stripping the trailing / (so we don't get double / in URLs).

Previewed e-mails on a server where the site root was not / and saw

that the generated URLs were correct.
Ran unit tests.