A shared wait budget across parallel workers: charge closed wall-clock time once, not each worker's sleeps

Pattern bug in a job runner whose workers wait for an external lock (a file must be absent at two checks N seconds apart) under a total wait budget. Each worker added its own sleep interval to one shared counter. That includes the mandatory second check after an open gate, and every concurrent waiter adds its own. Result: the budget runs out with the lock never taken (starts x spacing exceeds it), and while P workers wait together the budget drains P times faster than the wall clock. A small mock run showed it: lock file never created, budget smaller than starts x spacing, and the last job still ended with 'budget spent'. Fix: keep one shared deadline, set the first time any worker sees the lock held (deadline = now + budget), and have workers give up when now > deadline. Don't charge the open-path confirmation wait. Add a test where the lock never appears and the budget is smaller than the sum of the confirmation waits; the run must still complete.