Richer, feature-complete demo provider using vendor tools (amd-smi / nvidia-smi / lm-sensors) #7

Context

The initial LinuxProvider (branch feat/linux-provider) reads hardware sensors dependency-free from /proc and /sys and emits the canonical sensor IDs the default instrument panel binds to:

• cpu.total.utilization (/proc/stat)
• cpu.clock (cpufreq/scaling_cur_freq)
• cpu.temperature (hwmon: k10temp Tctl / coretemp Package)
• gpu.total.utilization + gpu.temperature (amdgpu gpu_busy_percent + hwmon edge)
• memory.used.percent + memory.used

This is deliberately minimal and portable (no external binaries required). It proves the Providers → Store → Binding → Widgets → Renderer pipeline with real data.

Goal

Build a richer, more feature-complete demo/provider that uses vendor tooling where available, both to surface deeper metrics and to serve as worked examples for users writing their own providers/panels.

Possible scope

• GPU (vendor tools):
  • AMD via amd-smi (or rocm-smi): VRAM used/total, GPU + memory clocks, power draw (W), fan RPM, per-engine activity.
  • NVIDIA via nvidia-smi (--query-gpu=... --format=csv): utilization, temp, VRAM, power, clocks, fan.
• CPU/system (lm-sensors): richer thermal/voltage/fan coverage via sensors/libsensors beyond the raw hwmon scan.
• Graceful capability detection: detect which tools/sensors are present at runtime; degrade cleanly (provider health Degraded with a message) when a tool is missing — never hard-fail.
• Example panels that showcase the richer sensors (VRAM gauge, power readout, fan tiles), demonstrating multi-sensor/customized panels.
• Docs: a "writing a provider" example walkthrough using one of these integrations (ties into docs/development/provider-guidelines.md).

Non-goals (for this issue)

• Replacing the dependency-free sysfs LinuxProvider — vendor-tool integration should be additive/optional, gated on tool availability.
• Windows/macOS provider work (track separately).

Acceptance (rough)

• A richer provider (or an extension to LinuxProvider) exposes additional GPU/CPU sensors when amd-smi/nvidia-smi/lm-sensors are present, with clean fallback when they aren't.
• At least one example panel + a short provider-authoring doc demonstrating it.

Filed as a follow-up to the sysfs-first Linux sensor work.