A run can have several attempts, but docs/workflows.md only explained the first
one. It said a canceled attempt is kept and linked to runner.md for everything
else, so rerun and the five-attempt rule were things a reader learned from the run
page or the CLI reference.
This adds the missing paragraph where the queue is described: an attempt is recorded each time a runner starts the job, a vanished runner hands it back, five attempts is the cap, and Rerun queues another attempt of the same pinned job rather than a new run. It also says a queued or running run cannot be rerun, which is a rule the UI enforces but the page never stated.
Part of the documentation item in #138 §6. No schema or runtime change.