make it easier to troubleshoot flaky CAPO CI pipelines

Today when we have a flaky CAPO CI pipeline, if we want to look at what is happening live on a run, we need to:

• start such a run
• cancel the cleanup job
• update the Heat stack to have the periodic cleanup spare it

Here is what I would propose:

• introduce a delay-capo-ci-cleanup-on-failure gitlab label
• on an MR marked with this label:
  • if a pipeline is successful, run the cleanup job as usual
  • if the pipeline is failed, don't clean it up, and instead add the please-delay-cleanup on the Heat stack