Allow disabling rollback on activation failure

[philon123]

deploy-rs allows to use a flag --auto-rollback false, which will ignore any systemd errors while activating a deployment.

That can make sense in certain cases, for example when upgrading kubnernetes. Kubernetes relies on several unts that have racy dependencies to each other, so we need to ignore errors when restarting all units at once. The races resolve after a couple seconds.

Another example in our project is that services might break for reasons unrelated to the deployment. For example, some certificate managing service might fail because a DNS entry was changed. In this case, we may want to ignore this failing service becasue it's irrelevant to the deployment.

I suggest introducing the same --auto-rollback false flag in cachix deploy.