ci: side-effects from #33362 #35416

#33362 has had some side-effects for anyone locally running the CI:

A side-effect of this change is that if you're running the CI (locally), and a job fails, you now cannot run any another job, until removing the failed container (this wasn't previously the case, you could run other job, just not the same job, without removing the failed container):

It seems like this problem is actually worse than described, because now you can no longer run jobs simultaneously on the same host. The script sets up a global network, but after https://github.com/bitcoin/bitcoin/commit/75cf9708a05d08c3de5dffd91228444d942254b8 instead of containers just having dynamic IP addresses, they now have conflicting hardcoded static ip addresses, so can't be started at the same time. Errors look like: Error: IPAM error: requested ip address 1111:1111::5 is already allocated to container ID 0dc...

cc @vasild

Fixes I can think of:

• revert #33362 (and lose its test coverage)
• ~use (and cleanup!) per-container networks~
• use dynamic IPs and pass them in to tests (least favourite)

~I think I would tent towards using a network per-container. It will only be "messy" if we don't tidy them up properly. We don't currently clean up networks ever (and I ∴ always have two lurking in docker network ls), and perhaps we should be, anyway.~

actually I don't think 2) will work even, as the networks will still overlap :(