Unrelated defect surfaced by a real gsd-test failure while verifying
#3339 (test/3339-fold-state-model-profile diffs zero files this fix
touches -- confirmed via `git diff <this-branch> origin/next -- ...`
returning empty for this file, ship.md, and gsd-core/bin/lib/*.cjs).
Fixed inline per this repo's no-defer policy for defects found while
building, rather than deferred.
The test's jq() mock shelled out to a full new Node.js process for
every mocked `jq -r .field` call inside the extracted track_shipping
bash script. The "exhausts the polling bound" test hits up to 24 cold
Node spawns in one spawnSync call (6 polling iterations x 4 jq calls),
racing a hardcoded 10s timeout -- under CI contention this tipped over
on the linux-node22 lane specifically while linux-node24 had headroom,
producing spawnSync's signal-killed `status: null` instead of the
process's real exit code, asserted against the expected 0.
Replaced the Node-subprocess jq() mock with a pure-shell sed -nE field
extractor that never forks a process, eliminating the variable-cost
operation rather than just widening the timeout. Verified extraction
correctness against sample JSON (head/status/checks/review, including
empty-string and non-matching-field cases). Locally verified (scoped
per session policy): all 9 tests in the file pass; the previously-
marginal test dropped from ~617ms to ~198ms.
The underlying track_shipping polling logic in ship.md was confirmed
correct and untouched -- this was a test-fixture race, not a product
bug.
