All deployments should have already their cluster_id filled in on
creation. Legacy deployments will not be retried as:-

* Ci::Build#retry calls `Ci::RetryBuildService`
* Ci::Pipeline#retry
calls `Ci::RetryPipelineService` which also calls
`Ci::RetryBuildService`
* `Ci::RetryBuildService` will clone a build to retry

It is also impossibly to backfill Deployment#cluster_id from
Project#deployment_platform correctly as clusters could have been
deleted, added or altered in the intervening time.