Add support for zero-downtime updates

[LongLiveCHIEF]

Whenever a node is utilized as a deploy node, the gitlab-ctl reconfigure command should be run with an env of SKIP_POST_DEPLOYMENT_MIGRATIONS=true. This shouldn't be a permanent env variable, for the node, so it makes sense to add a conditional to the Exec['gitlab_reconfigure that sets the env => attribute whenever a node is identified by a deploy_node boolean variable for the module.

Additionally, the Service['gitlab-runsvdir'] service is not designed to be restarted, and thus should not subscribe to config/package changes unless the implementor explicitly has a reason to do so. gitlab-runsvdir is a supervisor for the bundle of gitlab services on the host, and gitlab-ctl reconfigure refreshes these services depending on the configs that changed.

If puppet sends a refresh signal to the service, it will cause gitlab services to shut down, and make it impossible to do zero downtime updates.

Zero Downtime Updates Doc: https://docs.gitlab.com/omnibus/update/README.html#zero-downtime-updates`

[baurmatt]

If you run Gitlab in a single server setup, you do not have a "deploy node" but you might need SKIP_POST_DEPLOYMENT_MIGRATIONS as well :)

[LongLiveCHIEF]

Yeah, that's why I worded it the way I did (not mentioning HA specifically). It's labeled with the HA Support Milestone merely because we can't really claim this module supports HA fully until this feature works.