CI: Understand the rate limit and figure out whether it's still a problem

This is part of %(OKR 2022Q3 - 4.3) - Faster CI.

In the past, we hit a rate-limit issue in the CI when pulling Docker images. It is unclear what exact limit we are reaching. We want to increase the number of jobs to parallelize the CI more, but we are worried that it could cause this rate limit to be reached, resulting in errors. So we need to understand what exactly is this rate limit.

@davdumas may have some links for jobs that reached the limit.

It is possible that the recent move to manual pipelines (#3617 (closed)) gave us more breathing room.

If we fail to reproduce and to find a convincing explanation, we may need to ask GitLab directly. Or we can try to move on to the next step (add more runners: #3620 (closed)). If we then do reach the limit, we can revert and investigate.