Adds a platform-wide maintenance switch controlled from the admin panel.
A single kernel.request subscriber (priority 6, after the firewall listener)
short-circuits every request with 503, so no controller has to check it and
all API clients — the admin SPA, nobat724_front and clinic-pro-tauri — are
covered at once.
- SiteConfig gains five maintenance_* keys; no entity change, no migration
- MaintenanceService caches the state in Redis for 30s and is fail-open:
a Redis or database failure never takes the site down by itself
- API responses reuse the BaseController::error() envelope with code
MAINTENANCE_MODE plus a Retry-After header; browsers get a self-contained
Twig page (inline CSS, noindex) that renders even mid-deploy
- Whitelist keeps /oauth/*, the login endpoints and /api/v1/admin/settings
reachable, otherwise an admin could neither sign in nor switch it back off
- Admin bypass falls back to decoding the Authorization JWT, because several
admin-panel endpoints sit in the public_endpoints firewall (security: false)
where no token is ever resolved and isGranted always returns false
- A kernel.exception handler at priority 20 covers routing 404/405 and
firewall 401, which are thrown before the request listener runs
- app:maintenance on|off|status is the escape hatch when the panel is down
Also removes a stray `APP_SECRET = ...` line from .env.dev: the spaces around
`=` are rejected by Symfony Dotenv, which made every console command and the
whole app fatal. The secret already lives in .env.local, as the comment above
that line instructs.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
