Fetching the API route directory is currently the responsibility of the
API client ("store") service. The "store" service makes authenticated
API calls and therefore depends on the "auth" service.

This means that we cannot use the API route directory or the page links
returned from `/api/links` in the auth service itself, as this would
introduce a circular dependency.

Factoring out the responsibility for fetching the `/api` and
`/api/links` endpoints into a separate service which does not use
authentication provides a way to resolve this problem.

It also makes testing some aspects of handling these endpoints, such as
caching and auto-retry if the HTTP request fails, a little easier.