GitDeduplicationService performs idempotent operations on deduplicated
projects.

Immediate configuration is not ideal for group and instance
level clusters as projects that may never be deployed would
still have Kubernetes namespaces and service accounts created
for them.

As of https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/25586
we now create only the resources that are required for the
project being deployed, at the time of deployment.

Properly handle LFS Batch API response in project import

Closes #61624

See merge request gitlab-org/gitlab-ce!28223

Pull mirrors would run the `Projects::AfterImportService`, which would
force a `git gc` each time it finished. This is overkill and not
necessary now that we have refs packed more frequently
(https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/27826).

Closes https://gitlab.com/gitlab-org/gitlab-ee/issues/11556

Project imports were failing with `undefined method each_with_object for
String` because the import was attempting to parse the LFS Batch API and
failing due to the fact that the Content-Type wasn't a supported format
(e.g. application/vnd.git-lfs+json instead of application/json). We now
parse the body as JSON.

Closes https://gitlab.com/gitlab-org/gitlab-ce/issues/61624

Fixes #60425

PackRefs is not an expensive gitaly call - we want to
call it more often (than as part of full `gc`) because
it helps to keep number of refs files small - too many
refs file may be a problem for deployments with
slow storage.

In order to make `LfsImportService` more reusable, we
need to extract the logic inside `ImportService` and
encapsulate it into the service.

This new table will be used to store the external_dashboard_url which
allows users to add a link to their external dashboards (ex Grafana)
to the Metrics dashboard.