Subsystem 4e: Per-device unreachable status (switches/alarms/storage, inline)

[HandyS11]

Subsystem 4e — Per-Device Unreachable Status

Adds a persisted, three-way per-device reachability axis to the three managed smart-device types — switches, alarms, storage monitors — surfaced inline on each device's embed.

DeviceReachability: Reachable · Removed (entity destroyed — not_found) · NoPrivilege (token lost building access — access_denied) · NoResponse (timeout / other).

How it works

• Map — ReachabilityMapping.FromResponse translates RustPlusErrorCode → DeviceReachability (sealed inside Features.Connections; the protocol enum never leaks past it).
• Seam — device reads now return DeviceReading/StorageReading (payload + reachability); switch actuation returns DeviceReachability.
• Detect — connect-prime publishes a DeviceReachabilityChangedEvent per device; a periodic PollReachabilityAsync loop publishes on change (ReachabilityPollInterval, default 5 min); switch actuation publishes on failure. Also fixes a latent prime bug where a removed switch/alarm was silently primed as "off".
• Persist — Reachability column on the three entities (migration DeviceReachability, additive, default Reachable).
• Surface — each feature relay persists + re-renders (ownership-filtered by entity id); renderers show the reason inline and gate controls (Removed/NoPrivilege disable; NoResponse stays enabled — likely transient). Switch actuation replies with the localized reason and updates the embed. EN/FR throughout.

Verification

• 739 tests across 16 assemblies green; build -warnaserror with 0 warnings; jb cleanupcode --profile=ReformatAndReorder clean; localization parity 248 → 260 (+12 EN/FR keys).

Notes

• Inline-only — no #unreachable-devices channel — per the catalog's C3/C4 "prefer inline status, keep a channel only if it earns it".
• Does not fully close subsystem 4 — switch groups (4d) remains unbuilt.
• Known follow-up (deferred, non-blocking): a storage monitor recovering reachability mid-session briefly re-renders "unreachable" with Refresh disabled until the next contents trigger — storage has no persisted contents to recover from, unlike switches/alarms (which recover from LastIsActive). The naive fix would wrongly enable Refresh during a real server-down, so the correct fix (relay re-fetches contents on recovery) is worth its own small slice.

🤖 Generated with Claude Code