Over time, the webapidocs code has changed to accommodate changes in

our API, and so has our test data. These changes have largely been done

without properly auditing the resulting documentation, and unfortunately

starting in Review Board 4, we ended up with a lot of bad, mostly-empty

payloads and broken docs due to new objects in the database for the DVCS

work that took precedence over other curated docs. This has further

worsened with the changes for Review Board 5.
This change redoes a lot of how the doc generation code decides which

objects to pick. It still has the introspection capabilities, but now

specific docs can provide values for URLs to allow examples to be based

on specific entries. That means we can continue to augment the

documentation database data without further regressions.
The 4.0 release also introduced features that the documentation is still

saying must be turned on for use. That's very misleading. We now only

show a notice if a feature is listed as off by default.
Several examples with bad data would show up, and some would not. Same

with links. We worked around this with options for the directive to turn

these off, but with the fixes in this release, we no longer need to do

that for most resources.
Bad assumptions that led to hiding documentation for resources without

models have been fixed.
Examples now include the HTTP status code, and no longer show the wrong

payload if the response is a redirect.
And finally, the example URL now uses https instead of http.

Generated the docs and carefully went through every single page. Made sure

all URLs and examples were correct.
This does have a counterpart involving new database data, which will go

in separately.