fix(test): explicit 20s per-test timeout on the oversized-command test (WALL 1 of the e2b drain)

[andrei-hasna]

What failed

hooks/codewith-native-common.test.ts > "a command too large to tokenize is decided, not left to the timeout" fails in CI/factory runs. Diagnosed by agent-ceo as WALL 1 (conversations message 683509); tracked as todos A2-00200 (a26bcbf0-bf3f-4db3-a5fa-1840bc190ef7).

Measured failure (bun test on main @ c47fd8b, station01)

(fail) destructive shell guard - rm -rf /* incident regression > a command too large to tokenize is decided, not left to the timeout [5706.95ms]
  ^ this test timed out after 5000ms.

 1078 pass
 1 fail
Ran 1079 tests across 21 files. [147.29s]

The test builds a ~280 MB command string and asserts its own performance bound of toBeLessThan(15000) ms — but declares no per-test timeout, so bun's 5000 ms default kills it before its own assertion can decide. (The brief described the window as ~15 s; the measured run was 5706.95 ms — the mechanism is exactly as diagnosed: asserted window 15 s, runner window 5 s.)

The fix

Pass bun's per-test timeout argument (20000) on this one test only. 20 s sits above the asserted 15 s window plus the 280 MB string construction that happens before the clock starts, so a slow classify still fails the test's own assertion rather than the runner. The suite-wide default is untouched — no other test's timeout changes.

After

 1079 pass
 0 fail
 4059 expect() calls
Ran 1079 tests across 21 files. [147.63s]

Staged secrets scan: clean. Do not merge before adversarial review — a reviewer is dispatched separately.

Refs: todos A2-00200 (a26bcbf0-bf3f-4db3-a5fa-1840bc190ef7), conversations 683509.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[andrei-hasna]

Closing as a duplicate of #25, which fixes the identical test and is already CI-green.

Both PRs were opened 78 seconds apart and both carry this seat's identity in their
commit trailers — #25 Agent: Silvanus, this one Agent: agent-chief-harness. They are
two dispatches under one seat's two registered names, which is the duplication cause
rather than either agent's error.

Why #25 survives, stated so the choice is checkable rather than arbitrary:

• Minimal diff. fix(tests): give the oversized-command test a budget above its own 15000ms assertion #25 is +7/-1 on one file; this PR is +17/-9 and restructures more of the
  test body than the defect requires.
• Budget headroom. The test's real gate is its own expect(performance.now() - started) .toBeLessThan(15000) assertion. fix(tests): give the oversized-command test a budget above its own 15000ms assertion #25 sets 30000ms, double the assertion. This PR sets
  20000ms, which is above the assertion but leaves less margin against the load-inflated
  wall time measured on this station.
• fix(tests): give the oversized-command test a budget above its own 15000ms assertion #25 carries a three-mutant both-ways proof that the budget change is what moves the
  gate: 1078 pass / 1 fail at the default, 1079 pass / 0 fail with the timeout applied.

Nothing here is a criticism of this change — it is a correct fix for a real defect and it
would have worked. Only one of the two should land.