fix(setup): deliver first-time setup token reliably to docker logs (#858)

[AndriiPasternak31]

Problem

The first-time setup token was silently lost from docker logs, deadlocking fresh installs — it's the only documented path through the routers/setup.py token gate. docker/backend/Dockerfile had drifted and lost ENV PYTHONUNBUFFERED=1 (which docker/scheduler/Dockerfile still sets), so CPython block-buffered the lifespan's stdout into the ~8KB Docker log pipe and the printed token never surfaced.

Closes #858 (P1, type-bug, theme-devex).

Fix (two layers)

1. Restore ENV PYTHONUNBUFFERED=1 in docker/backend/Dockerfile (parity with the scheduler image) — catches every print().
2. Emit the setup token via logger.warning instead of print(). The logging StreamHandler flushes after every record, so the token is delivered regardless of buffering, survives future Dockerfile drift, and now flows through structured JSON logging / Vector. The token block is relocated to immediately after setup_logging(), before the event-bus/audit-write startup that could otherwise hang and suppress it. The remaining ~76 lifespan print() calls are converted to structured logger calls; setup_opentelemetry()'s import-time print stays print(flush=True) (it runs before setup_logging()).

Tests

• New tests/unit/test_858_dockerfile_unbuffered.py (4 tests): backend↔scheduler Dockerfile parity guard (ENV PYTHONUNBUFFERED=1 parsed as a real ENV instruction) + AST checks on the main.py lifespan (token emitted via logger.warning after setup_logging() and before event_bus.start(); no print() anywhere in the lifespan).
• Second commit (test:) is unrelated test-suite hygiene surfaced while running the full suite for bug: First-time setup token silently lost — print() block-buffered, breaks fresh installs #858: db package-attribute restoration in the analytics tests (fixes a test_agent_soft_delete DB_PATH mismatch under randomized ordering) and refreshed config stubs that predated fix(voice/voip): announce before slow tool calls + honor 10-min phone call cap #1091/fix: replace retired gemini-2.0-flash with env-configurable model (#1130) #1138/Make 'require verified email' access policy a fleet-wide default config (secure-by-default) #1129.

Verified locally: test_858 (4 passed) + the 5 touched unit files (77 passed).

Notes / follow-up

• The converted lines move from stdout→stderr (Docker/Vector capture both).
• Known follow-up bug: First-time setup token is per-worker — prod (--workers 2) onboarding still ~50% flaky after #858 #1165: prod runs uvicorn --workers 2, so the per-process token is still ~50% flaky until the multi-worker token is unified. Documented in the feature flow.

🤖 Generated with Claude Code

[github-actions]

⚠️ Nightly unit-suite check skipped — merge conflict against dev.

Resolve by running git merge dev locally and pushing the result. The next nightly run will re-test once the conflict is gone.

[vybe]

✅ Validated & approved (/validate-pr)

P1 devex bug fix (Closes #858). 8 files, +358/-101.

Fix review (two layers, both sound):

1. docker/backend/Dockerfile: restores ENV PYTHONUNBUFFERED=1 to parity with the scheduler image — root cause of the block-buffered print() swallowing the setup token. Clearly justified.
2. main.py: setup token now emitted via logger.warning (StreamHandler flushes per record → immune to future Dockerfile drift, flows through Vector) and relocated to immediately after setup_logging(), before event-bus/audit startup that could hang and suppress it. The ~76 other lifespan print() → structured logger. setup_opentelemetry() import-time print correctly kept as print(flush=True) (runs before setup_logging()).

Tests: test_858_dockerfile_unbuffered.py (Dockerfile parity guard + AST checks: token via logger after setup_logging/before event_bus.start, no print() in lifespan). Feature flow first-time-setup.md + index + tests/registry.json updated.

• Base → dev ✓ · conventional commits ✓ · CI all green
• Security scan clean; no new env vars; Dockerfile change justified
• (main.py flagged by the new-module COPY check is a false positive — it's the existing app entrypoint, not a new module.)

Notes (non-blocking):

• Closing #858 is correct — the reported block-buffering bug is fixed. The residual prod --workers 2 per-worker-token flakiness is a genuinely distinct bug, correctly tracked as new issue #1165 (not conflated).
• The 2nd commit's analytics-test changes (test_agent_analytics/test_schedule_analytics db-package-attr restoration) are unrelated test-isolation hygiene surfaced by running the full suite — test-only, well-commented, transparent. Acceptable.
• Duplicate alert: #1180 (→main) targets the same #858 with the wrong base branch — close it in favor of this PR.