Fix rollout failing when existing containers are in exited state

[Fikralaksana]

Summary
Fixes #20

When a service has containers in an exited state, docker rollout would attempt to restart the old (failed) container due to the --no-recreate flag, instead of creating a new container with the updated compose configuration.

This PR adds container state detection before starting services:

• Fetches container state via docker compose ps -a --format json
• If no containers exist at all, behaves as before (starts with --no-recreate)
• If existing containers are in exited state, runs up --detach without --no-recreate, allowing Docker to recreate the container with the updated config
• If containers are running normally, continues with the original --no-recreate behavior

Problem

1. Deploy v1 of a service — it crashes and enters Exited state
2. Fix the issue in docker-compose.yml (e.g., remove bad entrypoint)
3. Run docker rollout — it restarts the old broken container instead of creating a new one with the fix

[wowu]

Thanks for the contribution!

All changes to docker rollout must work in POSIX sh, work with docker compose v1 and v2, and don't expect any special tools like jq to be installed on the machine.