Keep retried phases in order with one worker
With execution.workers: 1, a phase waiting out its retry.delay could be overtaken by the next phase, which then ran on a bench the retried phase had not finished preparing.
- Fixed phase order under
workers: 1: every attempt of a retried phase, its retry delay included, now ends before the phase written after it starts, in setup, main and teardown alike. - Kept phases independent with more than one worker: a phase without
depends_onstill runs while another phase waits to retry. - Updated the workers and retry docs, and the
procedure.yamlschema, to describe this ordering.
