docs: correct the hardware and addressing facts this repo got wrong

[Gerrrt]

What changed

Nothing about the running lab — this is documentation catching up with what is
actually installed. A discovery pass against the live estate found several
recorded facts that were not true, and each correction here is measured rather
than inferred.

Where	Was	Is
docs/network.md, docs/architecture.md, README.md	shiva = the hypervisor at 10.0.30.10	shiva = the iLO BMC at 10.0.30.10; Saruman = the Proxmox host at 10.0.30.110
docs/hardware.md	oracle: i5-1235U, 32 GB, 2 TB SSD	Inspiron 15-3565: AMD A6-9200 (2 cores), 4 GB, 500 GB HDD
docs/hardware.md	ProLiant RAM and storage: —	2× Xeon E5-2680 v3 (48 threads), 128 GB, 2× 1 TB RAID 1
docs/hardware.md	prometheus: MacBook Pro (2012)	Retina 13" — 8 GB soldered, no built-in Ethernet, on a USB NIC
docs/security.md	"Two exceptions" to default-deny	Three. snmp-exporter polls 10.0.30.10 from VLAN 99
docs/security.md	Mains loss defended by a monitored UPS	mjolnir has no battery; the control does not exist

Also adds a Management section to docs/hardware.md — the BMC had no entry at
all — and closes the 10.0.30.10 roadmap item.

Why

These were not cosmetic. Two of them had become load-bearing:

• oracle's entry described a machine that does not exist. Planning work had
  already assumed a 32 GB host with Quick Sync was available for a service tier.
  It is a dual-core A6 with 4 GB.
• The 10.0.30.10 ambiguity was tracked in docs/roadmap.md and flagged in a
  > [!NOTE] in docs/network.md. It is settled: the SNMP target's
  hypervisor-bmc role label was correct all along, and the inventory was not.

The UPS is the one worth reading twice. mjolnir has no battery installed, and
its management card reports a full one regardless — 100% charge, 48.0 VDC, a
battery temperature, an hour of runtime, zero alarms. All derived, none
measured. The only honest signal it emits is the self-test, which returns
Refused — internal fault. Any alert rule keyed on charge, runtime or alarm
count therefore cannot fire, no matter how bad things get. The threat model no
longer claims a control that is not there.

docs/adr/0002 is deliberately left untouched. Per docs/adr/0001 ADRs are
immutable, and 0002 recorded the decision accurately at the time — the third
rule arrived later, with the monitoring stack. The correction belongs in
security.md, which is a living document.

Blast radius

• 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

Documentation only; no config, no compose, no rules touched.

One follow-up this surfaces rather than fixes: the iLO takes 10.0.30.10 by
DHCP, and prometheus/targets/snmp.yaml hard-codes it. Without a reservation, a
new lease silently breaks the target. Noted in the [!NOTE] block.

Verification

• make validate passes — all checks passed, including the SNMP inventory
  cross-check across targets, generator, render and example
• Deployed to the lab and confirmed working — n/a, documentation only. Every
  figure came from the live hosts: dmidecode, lsblk, free, lscpu,
  Prometheus /api/v1/targets, the iLO web UI and the APC NMC
• Docs updated (docs/, service README, or docs/roadmap.md)

---

Generated by Claude Code