test: harden flaky async transport and logger rotation tests

[Coldaine]

Why

Several tests use custom waitFor polling loops or real filesystem rotation with retry sleeps. Critique flagged flake risk in CI (especially Windows runners). Hardening now prevents intermittent red builds as the suite grows post–PR #48.

Hot spots

File	Pattern	Lines
shadow-agent/tests/capture/capture-transports.test.ts	Custom waitFor (4s timeout)	L22+, L157–278
shadow-agent/tests/capture/session-manager.test.ts	Duplicate waitFor helper	L28+, L92+
shadow-agent/tests/electron/start-main-process-privacy.test.ts	vi.waitFor	~L102
shadow-agent/tests/logger.test.ts	File rotation poll loop	L204–240

Acceptance criteria

• Extract shared waitFor to tests/helpers/ with fake-clock or event-based resolution where possible.
• Replace busy-poll transport tests with deterministic event emitters or vi.advanceTimersByTime where applicable.
• Logger rotation test uses temp dir + explicit size trigger without unbounded retry loop (or mark @flaky with ticket if unavoidable).
• Document CI timeout budget for transport integration tests.
• Zero flaky failures over 10 consecutive CI runs on main (or document exceptions).

References

• docs/plans/plan-testing-observability.md — determinism first
• PR #48

[kilo-code-bot]

🤖 Auto-Fix Update

I successfully implemented changes to fix this issue, but encountered an error while creating the pull request:

Branch 'session/agent_06a83e9d-8b07-4b4e-8b2b-efa1d225b317' does not exist on GitHub. The branch may not have been pushed yet. Please ensure the Cloud Agent successfully pushed the branch before creating a PR.

The changes are available on branch session/agent_06a83e9d-8b07-4b4e-8b2b-efa1d225b317.