We stopped calling the fallback so we can remove this now

We can now share project so that we don't have to load project twice.
Also, this extracts non-relevant logic out of Deployment.

Update DeploymentsController accordingly

This improves query performance of
MergeRequestsController#ci_environments_status a lot.

However this means old deployments that deployed to kubernetes clusters
with prometheus installations will no longer show performance metrics as
we cannot backfill cluster_id from deployment_platform with certainty
(clusters may be edited/added/deleted, which changes the results of
deployment_platform).

Now that KubernetesService can no longer be a DeploymentPlatform we can
remove all kubernetes client code and KubernetesService edge cases.

We still fallback to environment.deployment_platform until we can
backfill

We nullify when cluster is deleted as we want to keep the deployment
record around.

Add cluster as an optional association

We will have only cluster for deployments where the build deploys to a
kubernetes cluster

We will introduce #cluster at a later stage.
Also deprecate this method

Starting with Rails 5, belongs_to now adds a presence validation to the
association, and so as of shoulda-matchers 4.0.0 the belong_to matcher
follows suit and tests that this validation is there by setting the
association to nil and asserting that there are validation errors. This
exposed an error with the `validate_branches` method: we need to check
the source and target project exist.

https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/27687

This enables sending a chat message to Slack or Mattermost
  upon a successful, failed, or canceled deployment