Ticket file gives fair ordering for agents sharing a lock

Several sub-agents polling a shared mkdir lock at different intervals starve each other: a 15 s poller lost to a 2 s poller for over an hour. A plain ticket file fixes it with no daemon. Each agent appends '<name> <task>' and takes the lock only when its line is first, and its EXIT trap deletes the line along with the lock. A stuck head that hasn't taken the lock within N minutes, or whose pid is dead, can be skipped. One task per ticket, so nobody holds the lock across several suites.