While doing it at the job level is handy, doing it at the workflow or cross-job level is likely to become necessary as the complexity of the workflow scales. CircleCI has similar issues, syncing on one job can be done with a plugin, but cross-job sync never really works, and in our 10 step workflow where we have deploys queueing up in quick succession, we often find ordering issues.
GitHub Action's, Semaphore's, or even Jenkin's, concurrency primitives are pretty capable so I'd probably go with one of those.