fix(oss-health): refresh telemetry with real fleet data (#584)

## Summary
Refresh the OSS Health telemetry snapshot with the real fleet size. The
page showed ~50 clusters; the actual fleet is far larger.

## Why
`/api/overview` still undercounts ~3x. It counts
clusters/nodes/tenants/apps with an **instant** PromQL query at a single
timestamp, so VictoriaMetrics only returns series that are non-stale at
that instant (~5m), catching only the clusters that happened to report
in the last few minutes (~50) rather than those active over the period.
PR #5 fixed the current-month freeze and the `Tenant`-kind label but not
this windowing issue.

The Grafana telemetry-overview dashboard counts over the selected period
and shows the real numbers. This snapshot is taken from that source
(VictoriaMetrics).

## What
- **Clusters / Nodes / Tenants** are cumulative per period — distinct
clusters that reported during the month / quarter / year: **911 / 2339 /
2444**.
- **App counts** are taken from the 30-day window for all three periods,
to avoid transient load-test spikes inflating the quarter/year peaks
(`max_over_time` over a long window captures one-off bursts — e.g. a
FoundationDB load test).
- Applies the existing `hack/fetch_telemetry.py` whitelist/alias
filtering, so only canonical managed apps are listed.

## Notes
- Manual snapshot, pending a fix to the telemetry-server query (instant
→ windowed). The fetch cron stays paused (`workflow_dispatch` only)
until the server is the source of truth again.
- Companion server fix: cozystack/cozystack-telemetry-server (windowed
queries).

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **Chores**
* Updated telemetry data with relative time period windows (Last 30
days, Last 90 days, Last 12 months) for easier period reference.
* Refreshed health metrics snapshot and application data with latest
measurements.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Author: tym83