Non-blocking review findings from PR #351 (1)

[twistedmelonman]

Non-blocking concerns raised while reviewing PR #351 (perf(tests): stop an orphaned child holding the test's pipe open).

None of these blocked the merge. They are batched into one issue so a
review's findings stay one unit of attention rather than 1 separate
tracking issues; tick items off as they are addressed, and close this issue
when the list is done or the remaining items are judged not worth doing.

• Watchdog fallback kills a pid rather than a process group, leaving orphaned children (PR author comments in the diff, echoed by the claude automated review, bash/tests/test-pre-push-scan-timeout.sh:326-352 (IGNORER fixture comment) and the run_bounded implementation in git/hooks/pre-push)
  The diff's own comments state that the watchdog fallback path signals a pid rather than a process group, so a killed command's children survive. This PR works around the symptom in the test harness by redirecting the driver's output to /dev/null so an orphaned child no longer holds the command substitution's pipe open; it does not change the kill behavior itself. The comments attribute the underlying defect to dotfiles#350 and state the fixture's child shape was kept deliberately so that issue's eventual fix has something to assert against. Suggested action: confirm dotfiles#350 is still open and tracks the process-group kill, since this PR closes the test-slowness symptom but not the cause.

[twistedmelonman]

Duplicate of #350.

This issue's one item asks to confirm #350 is still open and tracks the process-group kill. It is, and it does — #350 carries the measured 4s-vs-60s comparison between the two run_bounded branches, the orphan proof, the CI-takes-the-fallback reasoning, and the fix direction with its race analysis.

Nothing here is untracked, so closing rather than carrying both.