Grade 3 checks. No tokens, passwords, or connection strings.

make check equivalent:
  .venv/Scripts/python.exe scripts/check_contract.py
  PASS: offline skeleton structure, local refs and Python syntax; application NOT tested.

Pinned image smoke, http://127.0.0.1:8080:
  .venv/Scripts/python.exe tests/smoke.py --base-url http://127.0.0.1:8080
  PASS: lab2 public API scenario; infrastructure and grading evidence remain separate.

Persistence on the same image. The public harness executes scripts/restart.sh.
On this Windows host CreateProcess cannot run a shell script (WinError 193), and the
only WSL distro is docker-desktop. The harness was therefore run inside docker:28-cli
with python3, the compose network, and the Docker socket. The hook was still
scripts/restart.sh. BASE_URL was http://api:8080.
  PASS: persistence preserved confirmed data and existing session.

Containers before and after that hook kept the same IDs (restart, not recreate):
  api        15bfccf88c7a
  postgres   cbe7139645a5
Volume still present: mtusi2026lab2-fork_pgdata

Second API start, same volume, migrations not re-inserted:
  before: migrations=1 users=14 messages=6
  after:  migrations=1 users=14 messages=6

Unknown member:
  .venv/Scripts/python.exe scripts/check_grade3.py
  PASS: unknown member returns 400 and leaves no chat row

Short TTL, separate from persistence. API recreated with SESSION_TTL_SECONDS=2,
then .env restored to 3600 and the API recreated again.
  PASS: same token got 401 after 3 seconds.

Postgres stopped, API left running:
  GET /health/live  200 {"status":"ok"}
  GET /health/ready 503 {"error":{"code":"unavailable","message":"Хранилище недоступно",...}}
  request_id in the body matched X-Request-Id. Body had no SQL.
  Postgres was started again; /health/ready returned 200.
