Geo: Container registry sync jobs should fail on failure to pull from the primary registry

Summary

When there is a severe configuration problem, the secondary's container repository sync job will mark the registry record as synced and exit successfully. Even though no replication will occur.

Steps to reproduce

For example, when the secondary's configs contain an incorrect primary registry API URL: gitlab_rails['geo_registry_replication_primary_api_url'] = 'somethingwrong'.

Or for example, when the secondary's configs contain an incorrect (local) registry API URL: gitlab_rails['registry_api_url'] = "somethingwrong".

Or for example, when the secondary's configs contain an incorrect token issuer: gitlab_rails['registry_issuer'] = "somethingwrong".

Then, when creating a new container repository, or when adding images or tags to an existing container repository, the changes do not replicate to the secondary's container registry.

Yet, the replication progress bar appears to be synced.

What is the current bug behavior?

Geo replication progress of container registry does not have failures when it is not working.

What is the expected correct behavior?

Geo replication progress of container registry should have failures when it is not working.

Relevant logs and/or screenshots

Possible fixes

it returns [] as the list of tags when it could have told me there was an error talking to registry instead