Maintenance/incident notice: in-app banner that works when the backend is down

[lilyshen0722]

Sam's ask: users should see a short notice when we're doing something disruptive, instead of a mysteriously broken app.

Design constraints that shape this:

• Deploys are zero-downtime since fix: keep Redis stable and make backend rollouts zero-downtime #676 — so this is NOT a deploy banner. It's for (a) unplanned incidents and (b) rare planned maintenance (migrations).
• It must render when the backend is unreachable — so it cannot be served by the backend. The frontend nginx pod survives backend outages independently.

Scope:

1. Automatic incident banner (no ops action needed): the SPA already knows when API calls fail — on repeated health/API failure, show a slim top banner: "Having a brief hiccup — reconnecting…" and clear it on recovery. Pairs with fix(socket): kalıcı pes etme yerine sınırsız yeniden bağlanma + backoff #685's socket reconnect. This covers 95% of real cases with zero operator work.
2. Planned-maintenance notice (operator-set): frontend polls a static /maintenance.json served by its own nginx (mounted via ConfigMap so ops can set a message + window with kubectl apply, no rebuild). When present: banner with the message; landing + login also render it.
3. Both banners: dismissible, non-blocking, v2-token styled.

Explicitly not in scope: a separate status-page product/domain. The GitHub Actions uptime probe (#674) remains the external truth; link "status" in the footer to its workflow page for now.