The old method of generating fake resource paths assumed that the first

model instance that matched resource.get_queryset(...) would be

suitable in all cases, i.e., this object would have all optional

relationships that corresspond to models. For example, it assumed that

the last modified review request would have all of the following:

- A review request draft

- File attachments

- Screenshots

- Reviews

- Draft file attachments

- Draft screenshots

- Draft reviews

- Review replies

- etc.
This is problematic when we want to add a new model instance (e.g., for

the in-process DVCs work which adds new models) because all these child

objects will have to be created for the last modified model instance.
Now, we use a backtracking algorithm to iterate over path-model pairs so

that if a model does not have the requisite child objects we can

continue onto the next instance. This way, we only need to add objects

to the database that correspond to new resources, instead of duplicating

objects from the previous parent object or recreating them entirely.

Built the docs.

With this patch applied, the docs in /r/10118/ build correctly.