I could have split this into several commits:
* DRY direct linking to images.
* Check S3 config before direct linking.
* Just check if service is public instead of relying on name.
Developers may copy a staging or production database or experiment with
S3 storage. But when the S3 config is missing then calling `service`
raises an ArgumentError due to a missing name.
Now we only try to call `service` if the S3 config is present.