I feel for you. I also wasted a lot of very painful hours trying to get it to work. I even had it working for a while before a docker update broke it -- turns out docker-compose's ipv6 support that many people relied on for years was a "bug" that they "fixed".

Ultimately I also gave up and now have a combination of port forwarding, nat64, and 10+ socat proxies in my docker-compose file. (Specifically, intranet->container and container->intranet are ipv6; but container->container is still ipv4)

More generally, I now try to keep my docker host as stock as possible. Whenever I'm reaching for daemon.json I just catch myself, take a step back, and say "what's the stupid but easy way to get this working".

Why? All you have to do is add a prefix to both the docker daemon and individual networks configured by compose. I do it and it's painless, never hit a bug. Just ensure the prefixes are at least /112.