Use loose foreign key for deployment_clusters.cluster_id

What does this MR do and why?

Replaces the foreign key constraint for deployment_clusters.cluster_id with a loose foreign key.

Some clusters have a very large number of associated deployments, which can prevent the cluster from being deleted as the statement timeout is reached while executing foreign key actions. A loose foreign key defers the foreign key action, allowing the cluster to be deleted without error.

Migration output

main: == 20220609195803 RemoveDeploymentClustersClusterIdFk: migrating ==============
main: -- transaction_open?()
main:    -> 0.0000s
main: -- foreign_keys(:deployment_clusters)
main:    -> 0.0024s
main: -- transaction_open?()
main:    -> 0.0000s
main: -- execute("LOCK TABLE clusters, deployment_clusters IN ACCESS EXCLUSIVE MODE")
main:    -> 0.0003s
main: -- remove_foreign_key(:deployment_clusters, :clusters, {:name=>"fk_rails_4e6243e120"})
main:    -> 0.0024s
main: == 20220609195803 RemoveDeploymentClustersClusterIdFk: migrated (0.0068s) =====

main: == 20220609195803 RemoveDeploymentClustersClusterIdFk: reverting ==============
main: -- transaction_open?()
main:    -> 0.0000s
main: -- foreign_keys(:deployment_clusters)
main:    -> 0.0044s
main: -- transaction_open?()
main:    -> 0.0000s
main: -- execute("ALTER TABLE deployment_clusters\nADD CONSTRAINT fk_rails_4e6243e120\nFOREIGN KEY (cluster_id)\nREFERENCES clusters (id)\nON DELETE CASCADE\nNOT VALID;\n")
main:    -> 0.0026s
main: -- execute("SET statement_timeout TO 0")
main:    -> 0.0001s
main: -- execute("ALTER TABLE deployment_clusters VALIDATE CONSTRAINT fk_rails_4e6243e120;")
main:    -> 0.0028s
main: -- execute("RESET statement_timeout")
main:    -> 0.0002s
main: == 20220609195803 RemoveDeploymentClustersClusterIdFk: reverted (0.0161s) =====

How to set up and validate locally

1. Create a legacy certificate-based cluster
2. Run a pipeline with the following job definition to create a deployment:

     deploy:
       stage: deploy
       environment: test
       script:
         - echo 'deploy'

3. Delete the cluster
4. Verify the job trace from the pipeline is viewable without error, and that the associated cluster_deployment record is (asynchronously) removed.

MR acceptance checklist

This checklist encourages us to confirm any changes have been analyzed to reduce risks in quality, performance, reliability, security, and maintainability.

• I have evaluated the MR acceptance checklist for this MR.

Related to #364575 (closed)