fix: handle pre-aborted fetch signals

[AysajanE]

Summary

Handle a caller signal that is already aborted when fetchWithTimeout starts, with a focused regression for the pre-aborted path.

Change

• client/src/__tests__/utils/fetchWithTimeout.test.ts
• client/src/utils/fetchWithTimeout.ts

Checks

• corepack pnpm@8.15.5 --dir client test:unit -- src/__tests__/utils/fetchWithTimeout.test.ts — passed in a network-isolated container.
• corepack pnpm@8.15.5 --dir client exec tsc --noEmit — passed in a network-isolated container.
• corepack pnpm@8.15.5 --dir client build:client — passed in a network-isolated container.

Fixes #17

---

AI assistance was used; I reviewed and own this change.

Verification

the repository's declared test command exited 0 on this exact head (5fbd2c8) in a network-off container, before this PR was opened.
Commands, environment, and hashes: receipt M-106 — checkable in ~30 seconds without trusting us.
This record covers Northset's own contribution; it is not maintainer verification.
Maintainers: request a separate private run for any PR in your queue — open a run request: https://github.com/northset-oss/verification-pilot/issues/new?template=request-a-run.yml or email oss@northset.ai.
For repositories already onboarded with Northset, adding northset-verify to a PR requests a run on that PR.

[michael-chipmates]

Good catch, this closes a real race with the new stop-voice control. Verified the guard and the test locally. Thanks!

[AysajanE]

Thank you for verifying it locally and for the merge — glad the guard closes that race.