docs: refactor docs structure and tighten assistant instruction policy

shrink root README into a landing page with a docs map and focused contributor guidance
add TV_POWER_RUNBOOK as the canonical TV power rollout and canary runbook
add CHANGELOG and move project history out of README-style docs
refactor src README into a developer-focused guide (architecture, runtime files, MQTT, debugging)
prune redundant older HDMI docs and keep a canonical HDMI_CEC_SETUP path
update copilot instructions to a high-signal policy format with strict anti-shadow-README design rules
align references across docs to current files, scripts, and TV power behavior

TV_POWER_HANDOFF_SERVER.md

@@ -0,0 +1,83 @@
+# Server Handoff: TV Power Coordination
+
+## Purpose
+Implement server-side MQTT power intent publishing so clients can keep TVs on across adjacent events and power off safely after schedules end.
+
+## Source of Truth
+- Shared full plan: TV_POWER_COORDINATION_TASKLIST.md
+
+## Scope (Server Team)
+- Scheduler-to-intent mapping
+- MQTT publishing semantics (retain, QoS, expiry)
+- Conflict handling (group vs client)
+- Observability for intent lifecycle
+
+## MQTT Contract (Server Responsibilities)
+
+### Topics
+- Primary (per-client): infoscreen/{client_id}/power/intent
+- Optional (group-level): infoscreen/groups/{group_id}/power/intent
+
+### Delivery Semantics
+- QoS: 1
+- retained: true
+- Always publish UTC timestamps (ISO 8601 with Z)
+
+### Intent Payload (v1)
+```json
+{
+  "schema_version": "1.0",
+  "intent_id": "uuid-or-monotonic-id",
+  "issued_at": "2026-03-31T12:00:00Z",
+  "expires_at": "2026-03-31T12:10:00Z",
+  "target": {
+    "client_id": "optional-if-group-topic",
+    "group_id": "optional"
+  },
+  "power": {
+    "desired_state": "on",
+    "reason": "event_window_active",
+    "grace_seconds": 30
+  },
+  "event_window": {
+    "start": "2026-03-31T12:00:00Z",
+    "end": "2026-03-31T13:00:00Z"
+  }
+}
+```
+
+## Required Behavior
+
+### Adjacent/Overlapping Events
+- Never publish an intermediate off intent when windows are contiguous/overlapping.
+- Maintain continuous desired_state=on coverage across adjacent windows.
+
+### Reconnect/Restart
+- On scheduler restart, republish effective retained intent.
+- On event edits/cancellations, replace retained intent with a fresh intent_id.
+
+### Conflict Policy
+- If both group and client intent exist: per-client overrides group.
+
+### Expiry Safety
+- expires_at must be set for every intent.
+- Server should avoid publishing already-expired intents.
+
+## Implementation Tasks
+1. Add scheduler mapping layer that computes effective desired_state per client timeline.
+2. Add intent publisher with retained QoS1 delivery.
+3. Generate unique intent_id for each semantic transition.
+4. Emit issued_at/expires_at and event_window consistently in UTC.
+5. Add group-vs-client precedence logic.
+6. Add logs/metrics for publish success, retained payload age, and transition count.
+7. Add integration tests for adjacent events and reconnect replay.
+
+## Acceptance Criteria
+1. Adjacent events do not create OFF gap intents.
+2. Fresh client receives retained intent after reconnect and gets correct desired state.
+3. Intent payloads are schema-valid, UTC-formatted, and include expiry.
+4. Publish logs and metrics allow intent timeline reconstruction.
+
+## Operational Notes
+- Keep intent publishing idempotent and deterministic.
+- Preserve backward compatibility while clients run in hybrid mode.