Fix send-integration e2e flake: teardown budget asymmetry (10s afterAll vs 120s beforeAll)

[waleedkadous]

Problem

send-integration.e2e.test.ts fails intermittently in the Tower Integration Tests CI job with Error: Hook timed out in 10000ms — after all 5 tests pass. It has hit three PRs in two days (#1283, #1290, #1301) plus multiple earlier branches, each time requiring a manual rerun.

Root cause

(Credit: the air-1288 builder's diagnosis on PR #1301.) The suite's afterAll declares an explicit 10_000ms budget, overriding the config's 300s hookTimeout, while beforeAll in the same suite gets 120_000 — an unintentional asymmetry. The teardown performs two unbounded deactivate fetches, stopServer's 2s SIGTERM wait, and three rmSync calls; on a loaded ubuntu runner that exceeds 10s.

Fix

• Raise the afterAll budget to 120_000, matching beforeAll.
• Bound the two deactivate fetches with AbortSignal.timeout(10_000) so a hung Tower can't consume the enlarged budget either — the teardown still completes and the existing .catch(() => {}) handles the abort.

Testing

Suite passes locally under the e2e config: 5/5 in 1.8s. No production code touched — one test file, +6/−3.

🤖 Generated with Claude Code

[waleedkadous]

Additional evidence from PR #1301's branch (air-1288 builder's tally): the send-integration teardown failed 3 of 4 first-attempt CI runs on that branch alone, passing on both re-runs. That hit rate is well past 'occasional flake' on loaded runners and supports the full fix (budget parity with beforeAll + bounded fetches) over a smaller bump.