UI Home Page: wire the Uptime stat to Prometheus metrics (implement the metric if missing)

[smunini]

Summary

The Uptime stat card on the UI Home Page (/ui) is a hardcoded placeholder. It always renders 99.98% / "last 30 days" regardless of how long the server has actually been running.

crates/ui/src/lib.rs:1375:

// Not part of the resource-count read path — placeholder until the bulk
// export job-state and availability read paths are wired.
export_jobs: "13".to_string(),
export_jobs_queued: 1,
uptime_percent: "99.98".to_string(),

The card should be driven by real data from the Prometheus metrics API rather than a constant.

Current state

• crates/ui/templates/pages/index.html:23-25 renders metrics.uptime_percent with the label card-uptime ("Uptime") and sub-label card-uptime-sub ("last 30 days") — see locales/{en,es,de}/main.ftl.
• helios-observability already exposes:
  • crates/observability/src/uptime.rs — uptime_seconds(), process start-time tracking.
  • crates/observability/src/metrics.rs — GET /metrics Prometheus exposition; sets an uptime_seconds gauge on each render.
• So a process uptime metric exists (uptime_seconds), but an availability percentage over a window (what the card currently claims) does not — that would require historical scrape data (a Prometheus server), not just the in-process gauge.

Scope of work

1. Hook the Home Page Uptime card up to real metrics instead of the hardcoded string.
2. Decide (and then implement) what the card actually shows:
   • Option A (simplest, self-contained): show process uptime duration from uptime_seconds (e.g. "3d 4h") and change the sub-label away from "last 30 days". This needs no external Prometheus server. Note crates/ui/src/i18n.rs already has a health-uptime string ("Uptime: {duration}") that may be reusable.
   • Option B: query an external Prometheus for an availability percentage over a window (e.g. avg_over_time(up{...}[30d])), keeping the "% / last 30 days" presentation. Requires configuration for the Prometheus base URL and a graceful "unavailable" state when it isn't configured.
3. If a needed metric isn't available, implement it in helios-observability and export it via /metrics.
4. Handle the unavailable/unknown state in the UI (don't render a fabricated number).
5. Update the affected locale strings in locales/en|es|de/main.ftl if the sub-label changes.
6. Add tests covering the new read path (crates/ui Rust tests; Playwright if the rendered text changes).

Notes / constraints

• /metrics is public and unauthenticated — per the comment in crates/observability/src/metrics.rs, tenant must never appear as a metric label. Don't add per-tenant data to satisfy this card.
• uptime_seconds in a clustered deployment behind a load balancer reflects only the process that served the scrape (see the doc comment in crates/observability/src/uptime.rs) — worth calling out in the UI copy or the chosen approach.
• The sibling placeholders on the same card row (export_jobs, export_jobs_queued) are out of scope here unless trivially covered.

Acceptance criteria

• Home Page Uptime card reflects real server data, not a constant.
• The card's label/sub-label accurately describes what is shown.
• A sensible fallback is rendered when the metric can't be obtained.
• Tests cover the new path.