[PRIORITY] Do not hard error if containers are gone

[apostasie]

This is another instance of #3167 (see ticket for rationale - in a shell: do not assume containers are still there after a call to client.Containers...)

In that specific case, it will make create fail if another unrelated container got removed in a racy way.

It is very likely responsible for a large number of failures on the CI, and definitely strikes more with parallelization.

This very likely will address #3092, #3186, and possibly a number of others.

@AkihiroSuda @fahedouch if you are around - would appreciate a quick merge on this, preferably before other QA/CI PRs so that I can rebase and see what problems are left (specifically for the IPFS one).

Of course, I will rebase #3189 as well ASAP which should give us a good hint.

Finally, if this is as bad as I think it is, merging this should give us a good speed boost on the CI (as a lot of "retries" will no longer be necessary).

Thanks a lot!

[apostasie]

Note: in the future, when we will be in a better position wrt CI, I will add tests to stress concurrency to ensure these problem patterns do not get reintroduced in the codebase.

[apostasie]

Note: failure seems entirely unrelated (systemd socket something)

[AkihiroSuda]

Thanks

[apostasie]

Last run failure on integ-rootless is IPFS - these should get fixed soon with #3181