feat: per-CPU temperature sensors over MQTT/Home Assistant

[sethpjohnson]

What

Breaks out per-CPU-socket temperatures over MQTT so Home Assistant shows one temperature sensor per physical CPU, not just a single aggregate. On a two-socket R730 (e.g. bandit) this turns 1 CPU sensor into 3: CPU 1, CPU 2, and the existing max.

Mirrors the per-GPU implementation shipped in v1.2.0.

Changes (all in internal/mqtt/)

• State payload: new cpus array — one {index, temp} per socket, sourced from status.CPU.Temps (already read; only .Max was published before).
• Per-CPU discovery: cpuN_temp HA sensors (0-based ids cpu0_temp/cpu1_temp, 1-based labels "CPU 1/2 Temperature"), device_class: temperature, °C, cold-start-safe default('') value_templates indexing value_json.cpus[i].temp.
• Lazy publish: publishNewCPUDiscovery mirrors publishNewGPUDiscovery — mutex-guarded, gated on a successful state publish, cleared on reconnect (handles connect-before-first-read).
• Aggregate preserved: cpu_temp (max) sensor unchanged in id/value key, relabeled "CPU Temperature (Max)" for parity with the GPU max sensor. Fan logic and existing HA history depend on it.
• Scales to N sockets; nothing hardcoded. internal/controller untouched — control loop stays MQTT-unaware.

Test evidence

• go vet ./... clean · CGO_ENABLED=1 go build ./cmd/controller ok · go test -race -count=1 ./... all 8 packages pass.
• New/extended tests: buildStatePayload cpus population (2 & N sockets, nil when no reading), cpuDeviceSpecs ids/labels/device_class/template, lazy discovery once-per-index + reconnect re-announce, key-correspondence guard for cpus[].temp, aggregate relabel.
• Acceptance: driven against a real mosquitto broker in --demo — observed "cpus":[{"index":0,"temp":41},{"index":1,"temp":42}], both cpuN_temp/config discovery messages with correct topics/labels, aggregate relabeled, values updating per tick, all confirmed as retained messages.

Task: dex yqkz8t1k

🤖 Generated with Claude Code