mesh: LightOverrideMsg — command a single light across devices

[bennotk]

Summary

Stacked on #462 (base branch claude/issue-457-override) — merge that one first; GitHub retargets this PR to main automatically when the base branch is deleted.

• New additive MsgType::LightOverride (37) with LightOverrideMsg: targetMac[6] + lightIndex (addressing mirrors SetGroupMsg, target filtering after parse), full LightConfig snapshot, durationMs (0 = stays until displaced), and a clear flag so any device — later a graph node — can return a light to its group without supplying a snapshot
• Receiver wiring in main.cpp: a targeted device stores/clears the override through the exact same config,web: per-light override with last-command-wins arbitration #457 arbitration path as the local REST surface (LightOverrides + applyEffectiveLight + peers push), so concurrent senders resolve by receive order — last one wins; non-targets ignore the message entirely
• MeshManager::broadcastLightOverride() takes a fully-built message — snapshot construction deliberately stays with the sender (the set-light button action, actions,web: set-light button action targeting a light on any device #459, per the rework plan)
• Inventory row in docs/mesh-compatibility.md: fixed-size, len >= sizeof(LightOverrideMsg), same-firmware wire policy, no existing message touched
• No mock/UI changes needed: the dashboard's "Overridden" indicator from lights,web: per-light override with last-command-wins arbitration #462 reads the light's override state regardless of origin, so a mesh-arriving override shows up identically via the existing peers push

Closes #458

Test plan

• npm run lint and npm test (44/44) pass — no JS touched, run as regression
• clang-format --dry-run --Werror clean on all changed src/ files
• On pr- OTA builds on two devices (A sends, B receives — needs a sender, which lands with the actions,web: set-light button action targeting a light on any device #459 button action; until then reviewable as message + receiver wiring): a broadcast targeting B overrides exactly the addressed light on B while other devices update nothing, clear returns it to the group from any sender, B's dashboard flips its indicator both ways

🤖 Generated with Claude Code

https://claude.ai/code/session_01QCqsYkCrhRHLoWZFuDTWjh

---

Generated by Claude Code

[bennotk]

Same infrastructure-level CI failure as on #462: every job fails within ~2 seconds with no runner assigned (runner_name empty), no logs, no output — GitHub isn't provisioning runners (most likely the account's Actions minutes/spending limit). Nothing to fix in this branch; lint, tests (44/44), and clang-format are green locally. I'll re-run the workflows on my next check-in.

---

Generated by Claude Code

[bennotk]

CI is green after rebasing onto the renamed project's main and onto #462's fixed head — all 10 checks pass, including both firmware builds and the published pr-463 build.

Still stacked on #462 (base branch claude/issue-457-override); merge that one first and GitHub retargets this PR to main automatically.

---

Generated by Claude Code