Preserve teardown completion after an intermittent timeout race #1376

The existing slow teardown acceptance test still fails intermittently on macOS Bash 3.2.57. A clean runtime from main at f33a9ac2c87ca7e181d76a4b3e65eddd3de8ea7a reproduced the failure during a repeated focused run. The timeout message is correct, but the marker written after tear_down sleeps for one second is absent.

./bashunit --simple --parallel --filter waits_for_a_slow_tear_down --repeat 30 tests/acceptance/bashunit_timeout_test.sh

The clean-main run failed after 14.39 seconds. Other repeated runs passed. The same check also failed during the full standalone suite while validating #1370. The test and timeout execution path are unchanged by that provider work, and the failure was reproduced independently of it. The root cause is not established.

#1344 replaced the short fixed grace with polling for up to two seconds. Investigate why a one-second teardown can still miss its marker without making a hanging teardown unbounded.

• Identify the timeout or teardown race with a focused regression.
• A timed-out body runs its teardown once and permits a one-second cleanup to finish.
• A hanging cleanup remains bounded, and watchdog children do not delay output capture.
• Validate the source and standalone runner on macOS Bash 3.2 and actual Bash 3.0, including repeated and parallel runs.