Support worker autoscaling

[gravesm]

Adding autoscaling to ECS for the worker service requires a few changes
to how the cluster is redeployed. Previously, everything was stopped by
setting the desiredCount to 0 and then started back up by setting the
desiredCount to 1. With ECS controlling the scaling for the worker
service it's no longer possible to rely on the desiredCount of the
worker service remaining at 0 until things are ready to be started
again.

The change here sets the desiredCount of the scheduler to 0 first, then
does a --force-new-deployment for the web and worker services, then
finally does a --force-new-deployment for the scheduler. While
redeploying becomes more complicated, there's no need anymore to track
the current worker count. We can just set it to 1 and let ECS scale as
needed.

I've also added stop-scheduler and start-scheduler commands which can be
used in conjunction with deploying a new service definition with
Terraform.

How can a reviewer see these changes?

You can run:

$ pipenv run workflow redeploy

and verify that the command completes without blowing up.

Reviewer Checklist

• The commit message is clear and follows our guidelines
• There are tests covering any new functionality
• The documentation has been updated if necessary
• The changes, if applicable, have been verified