fix(observability): show alerts with a table, not the alert list panel

[Gerrrt]

Fixes #37, whose panels rendered nothing. Verified on screen this time.

What was wrong

Three attempts, each wrong differently:

1. Alertmanager datasource. Chosen so silences would show. But alertlist
   reads alert rules, and an Alertmanager datasource has none — nothing to
   enumerate.
2. Prometheus/Loki by uid. UnifiedAlertList.tsx matches
   dataSourceName === options.datasource — the datasource name, not the
   uid. Nothing matched.
3. Name corrected. Still No alerts matching filters — and still empty with
   the label filter removed entirely. That bisect settled it: the panel source
   says this should work, it does not, and the fault is not visible server-side.

Every one of those passed check_dashboards.py, validate.sh, and my own API
checks. None of those touch what the panel actually draws, which is the whole
lesson here.

What this does instead

Plain table panels over the ALERTS series Prometheus exposes for its own
rules:

ALERTS{component="network",alertstate=~"firing|pending"}

The point is that this is an ordinary PromQL query, so the exact expression each
panel runs was verified through Grafana’s own datasource proxy before asking
anyone to look at a screen.

Columns: Alert, State, Category, Device, Instance, Severity, VLAN — with
severity and state colour-mapped.

Two deliberate costs

Silences are not reflected. ALERTS is Prometheus’s view; silences live in
Alertmanager, so a silenced alert still shows. That was the original argument
for the Alertmanager datasource, given up for a panel that works.

The Logs dashboard loses its panel. Loki’s 8 rules are evaluated by Loki’s
ruler and go straight to Alertmanager — they never enter Prometheus ALERTS, so
component="logs" would be permanently empty. A panel that always reads clear
implies there are no log alerts, which is worse than no panel.

Both are worth revisiting if alertGroups ever ships as a dashboard panel in
Grafana OSS, or if the alertlist issue is understood.

Also reverted

The check_dashboards.py alertlist exemption from #37 — no alertlist panel
remains, and it documented behaviour that does not work here.

Blast radius

Dashboard JSON only, applied with make reload.

• No change to network segmentation or firewall rules
• No new port published to a VLAN that could not already reach the service
• No credential added outside secrets/*.sops.yaml

Verification

The panel’s exact query, through Grafana’s proxy:

ALERTS{component="network",alertstate=~"firing|pending"}
-> Alert=IloBatteryCondition State=firing Severity=warning Device=shiva Instance=10.0.30.10

Confirmed rendering on screen — the table draws all seven columns and the
firing row. That is the check the previous three attempts lacked.

./scripts/validate.sh passes — 5 dashboards OK, 84 panels, 83 PromQL
expressions.

• make validate passes
• Deployed to the lab and confirmed working
• Docs updated (panel descriptions carry the rationale and the silence caveat)

Refs #36