DeviceManager: order-based discovery (extracted from #865)

[Christian-Holbrook]

What

Extracts just the orders system from #865, re-grafted onto current master (which has since gained Roku Cloud Emulator support — the source of #865's conflicts).

Triggers no longer scan/health-check directly. Instead they submit orders into per-type pending sets, and visible views fulfill them:

• Order types: broadcast (SSDP scan) and reconcile (health-check all), each with a reason (startup, network, sleep, refresh-clicked, config-changed, unhealthy-device, stale).
• Reasons accumulate and dedupe; one fulfillment satisfies every queued reason; take is atomic per type.
• Nothing hits the network while no view is visible — orders queue until the Devices panel or device picker opens (or is already open).
• Live orders are fulfilled immediately by visible views via the new order-submitted event (replacing scanNeeded-changed); timer-driven stale orders are excluded from live fulfillment and are staleness-gated inside broadcast().
• The unhealthy-device rescan is now a rate-limited order (1/min), terminating the old scan → fail → scan feedback loop.
• Removed: scanNeeded flag, setScanNeeded, refresh(), scan(), discoverAll().

What this deliberately does NOT include from #865

The freshness rewrite (ensureDeviceFresh, peekDevice, lastCheckedByIp trust windows, getDeviceInfo consolidation, hidden-view lazy reads). reconcile() drives the existing healthCheckAllDevices(force) machinery unchanged; healthCheckDevice keeps its current signature.

Testing

• Ported the orders test suites from DeviceManager rewrite: order-based discovery #865 (Orders unit tests, submit/fulfill/stale-timer/broadcast/reconcile, picker order flow, view-provider fulfillment) and adapted them to master's API.
• Full suite: 1308 passing / 1 failing — the failing BaseWebviewViewProvider timeout also fails on clean master (pre-existing, unrelated).

🤖 Generated with Claude Code