per-unit timeout not effective, overloading CI because of long runs

A mistake slipped in !7039 (merged): the removal of the --unit-timeout parameter was thought useless given that all Kustomizations/HelmReleases now have a sylvactl/unitTimeout annotations.

But this parameter is necessary to keep, or else sylvactl does no per-unit timeout processing, and ignores the sylvactl/unitTimeout parameter.

As a result, we've observed many CI deployment pipelines running until the global CI job timeout (nearly 3 hours), severely impacting the time in which we get feedback, and increasing the pending job queue time on runners limiting the number of parallel jobs (in particular capo).