Reduce downtime window when deploy_with_maintenance

[dacook]

Description

# This playbook is meant for deployments that introduce big changes and thus
# require enabling maintenance mode and refreshing the cache.

However, it shuts down all services for the entire deploy role, which can sometimes take a long time (I think 20 mins?)

The first half is mostly setup (preparing ruby, gems, node, etc). So I think we only need to shut down for the second half of that role, which could save a lot of time.

Acceptance Criteria & Tests

Run the playbook before, and time how long the site is down for (or just review the playbook log)

Run it after. The site should be usable up until the maintenance mode, and again after. It should be unusable for a shorter time.

[dacook]

I just checked one example from a deployment for a single server.
It took 5 mins, but the first three mins were setup, and it only took two minutes after task check database status.