fix(e2e): unblock nightly-deploy-moddable cell

[ottovlotto]

Summary

Three test-infra fixes for the first-night failure on the moddable cell (issue #202). The CLI itself worked correctly on the failing nightly — exit 0, deploy completed, registry stamped. The cell failed because:

• vitest hung ~30 s after junit was written, killed by the action runner → exit 1 → "attempt 1 failed";
• nick-fields/retry re-ran the setup; attempt 2's gh repo create paritytech/e2e-cli-moddable-<run_id> collided with attempt 1's leftover repo ("Name already exists") → real failure.

1. Drop the un-unref'd 30 s setTimeout that caused the vitest hang

src/utils/connection.ts::timeoutAfter and the duplicate in e2e/cli/helpers/chain.ts::getTestClient both create a setTimeout(..., 30_000) inside Promise.race. When connectPaseo() wins the race, the loser timer is not cancelled — it stays pending and keeps Node's event loop alive until it fires. The CLI's scheduleHardExit() in process-guard.ts papers over this in production; the vitest harness has no equivalent and was hanging past its 5 s teardownTimeout, surfacing as:

close timed out after 5000ms
Tests closed successfully but something prevents Vite server from exiting

Fix: .unref() on the timer in connection.ts. Removes the redundant wrapper in chain.ts (it was layering a second 30 s timeout on top of the one already inside getConnection()).

2. Make setupModdableFixture retry-safe

Probe gh repo view paritytech/<repoName> before gh repo create. If the repo already exists (retry within the same run, or leftover from a crashed earlier run), reuse it: git remote add origin … && git push -u origin main --force. Mirrors what the CLI itself logs on re-run (using existing origin (...)). Force-push is safe — fixture content is identical between attempts.

gh auth setup-git is invoked before the raw git push so HTTPS auth via GH_TOKEN flows through gh's credential helper (which gh repo create --push does implicitly).

3. Cleanup post-step in the workflow

if: always() && matrix.cell == 'nightly-deploy-moddable' step that runs gh repo delete paritytech/e2e-cli-moddable-${{ github.run_id }} --yes || true after artefacts are uploaded. Runs on success, failure, and cancellation. The weekly e2e-cleanup.yml cron stays as the backstop, but per-run cleanup means we don't accumulate one orphan per failed nightly.

Test plan

• pnpm format:check — clean
• pnpm lint:license — clean
• pnpm build — clean
• pnpm test — all unit tests pass (one telemetry test is a known flake under suite-wide contention, passes in isolation)
• nightly-deploy-moddable cell passes on the first attempt (verified via workflow_dispatch trigger after merge)
• Subsequent scheduled nightly is green and Nightly E2E failure: 2026-05-22 #202 auto-closes

Closes #202

[github-actions]

Dev build ready — try this branch:

curl -fsSL https://raw.githubusercontent.com/paritytech/playground-cli/main/install.sh | VERSION=dev/becca/e2e-moddable-cell-fixes bash

[github-actions]

E2E Test Pass · ✅ PASS

Tag: e2e-ci-pr · Branch: becca/e2e-moddable-cell-fixes · Commit: 0836c15 · Run logs

Cell	Result	Time
pr-deploy-cdm	✅ PASS	2m06s
pr-init-session	✅ PASS	1m45s
pr-deploy-foundry	✅ PASS	0m38s
pr-deploy-frontend	✅ PASS	3m30s
pr-install	✅ PASS	0m43s
pr-preflight	✅ PASS	1m26s
pr-mod	✅ PASS	2m02s
${{ matrix.cell }}	⏭️ SKIP	0m00s
${{ matrix.cell }}	⏭️ SKIP	0m00s

Sentry traces: view spans for this run

[ottovlotto]

Verified the moddable fix via workflow_dispatch on this branch (run 26298463347):

• nightly-deploy-moddable ✅ passed in 111 s on first attempt (no retry — "Surface failure detail" step skipped, confirming the test cell exited cleanly within vitest's 5 s teardown budget).
• Cleanup post-step ran successfully — the per-run repo was deleted.
• All other nightly-only cells (nightly-mod-miss, nightly-deploy-hardhat, nightly-deploy-multi, nightly-chaos-rpc, nightly-chaos-sigint, nightly-diagnostic, nightly-rejections, Init cold-start smoke test) also green.

pr-init-session and pr-install were cancelled within ~20–50 s of starting ("The operation was canceled" — GH runner cancellation, not a test failure). The pull_request CI on this same branch run 26293100972 had all pr-* cells green, so the cancellation here was environmental.