It has gotten a bit better lately, but it's still a pain.

Just recently the docker-compose stack from the current main branch would not start properly, because someone committed a faulty health check. Why does this make it to the main branch, does nobody there review PR?

Starting a new stack is nothing compared to maintaining it. Upgrading to newer images requires carefully checking which environment variables suddenly appeared in new versions or maybe were renamed. Upgrades never really went absolutely smooth for me in the past.