EH/MU/FR-C1: live sensor-health KPIs, stale detection, refresh interval

[Mustafa-1089]

Summary

Turns the Sensor Health dashboard (FR-C1) from a static shell into a live view. The KPI cards, status pills, stale indicators, and refresh interval are now all driven by real data from the backend's /sensors/updates endpoint instead of hardcoded placeholder values.

Why

Before this change the page had the right structure but wasn't meeting the FR-C1 acceptance criteria:

The three summary cards showed fixed text ("Live", "MQTT / Seeded", "15s") that didn't reflect the actual sensors, so the KPIs and the table could say different things.
Nothing surfaced stale or offline sensors, the backend already derives a Degraded/Offline status, but the frontend had no pill, filter, or visual treatment for it.
The "Last Audio" column read a field (item.lastAudio) that the backend doesn't return, so it was always blank.
The refresh interval was hardcoded in three separate places that could drift apart.

What changed

Live KPIs: Replaced the 3 static cards with 5 (Total / Online / Degraded / Offline / Low Battery), computed from the same array that fills the table, so the cards can never disagree with the rows.
Stale detection: Degraded and Offline rows are now visually highlighted, each status pill shows a "seen Xm ago" subtext from lastSeenMinutesAgo, and a "Degraded" option was added to the status filter.
Refresh interval: Introduced a single REFRESH_MS constant driving both the auto-refresh timer and the displayed "auto-refresh every 15s" label, one source of truth.
Last Audio column: Now maps the real lastAudioMinutesAgo field instead of a non-existent one.
Low-battery highlight: Keys off the backend's derived Low Battery status rather than a hardcoded <20 threshold, so frontend and backend agree.

Testing

Verified against the real backend, not just a syntax check:

Seeded 5 of the existing topology nodes with health fields; GET :9000/sensors/updates returned all 14 nodes with the expected mix (2 Online, 1 Degraded, 1 Offline, 1 Low Battery, rest Offline/null).
Confirmed KPI counts sum correctly to the total and match the table.
Loaded the page and confirmed the new cards, filter, and layout render.

Not in scope / known follow-ups

• The page fetches through the HMI proxy on :3000, which currently returns empty while the backend on :9000 returns data correctly. That's a separate HMI proxy/routing issue (FR-D1), not part of this change.
• CPU / RAM / Disk / Uptime / GPS columns show "--" pending a backend telemetry source; the backend derivation doesn't return them yet.

[Vamshi-Gollapelly]

I went through the changes and there are a couple of things that should be addressed before this is ready to merge. I left inline comments on the battery/status handling and KPI counting logic, as both could produce incorrect results in some scenarios. Once those are updated and verified, I'm happy to review it again.

[Vamshi-Gollapelly]

The batteryPct < 20 check above still runs before any status check, so a Degraded/Offline sensor with low battery will show "Low Battery" here even though the KPI card and row highlight correctly show Degraded/Offline. Since status is the source of truth everywhere else in this PR, can you fold it in here too and drop the raw threshold branch?
```js
function pillHtml(status) {
const s = String(status || "").trim();
if (s === "Online" || s === "Success") return <span class="pill pill-success">${s}</span>;
if (s === "Degraded" || s === "Low Battery") return <span class="pill pill-warning">${s}</span>;
if (s === "Offline" || s === "Failed") return <span class="pill pill-danger">${s}</span>;
return <span class="pill pill-warning">${s || "Unknown"}</span>;
}
```

[Mustafa-1089]

The battery check was overriding status. I've dropped the raw batteryPct threshold from pillHtml and it now trusts the backend status everywhere (pill, row highlight, and the filter). So a Degraded or Offline sensor with low battery now shows its correct status. Pushed in the latest commit.

[Vamshi-Gollapelly]

When backend status is something other than Online/Degraded/Offline/Low Battery (your testing notes mention nodes coming back Offline/null), updateSensorKpis counts it toward Total but doesn't add it to any bucket. Can you confirm the backend always returns one of those four values, or should this have a fallback bucket so the cards can't silently under-count?

[Mustafa-1089]

The backend derives exactly four statuses (Online / Degraded / Offline / Low Battery), so those are the full set today, but you are right that the cards shouldn't silently under-count if that ever changes. I've added an other fallback bucket so every sensor is counted and the buckets always reconcile with the total. Pushed in the latest commit.

[AnDo081105]

The interval is inside #last-updated-at, but updateLastUpdated() replaces the parent’s entire textContent at script.js (line 209). Nothing assigns REFRESH_MS to the span, so the required visible interval changes from -- to nothing

[AnDo081105]

The backend overwrites Offline/Degraded with Low Battery when the battery is below its threshold.The frontend then adds the stale treatment only for Degraded or Offline. So a low-battery sensor whose heartbeat is old or missing appears only as low battery?