CUB-204: wire WS client as primary, REST poller as fallback

- Rename GatewayURL/GatewayPollInterval → GatewayRestURL/GatewayRestPollInterval
- Change Docker-aware defaults (host.docker.internal instead of localhost)
- Client.Start() waits for WS readiness (30s timeout), falls back to REST
- Client.SetWSClient()/MarkWSReady() for WS→REST coordination
- WSClient.SetRESTClient() so WS notifies REST on successful handshake
- main.go wires both clients: WS primary, REST fallback with cross-references
- .env.example documents WS_GATEWAY_URL, GATEWAY_TOKEN, REST fallback vars
- docker-compose.yml adds WS_GATEWAY_URL and GATEWAY_TOKEN env vars
- reference/CONTROL_CENTER_CONTEXT.md documents architecture and startup sequence

reference/CONTROL_CENTER_CONTEXT.md

@@ -0,0 +1,46 @@
+# Control Center — Architecture Context
+
+## Current State
+
+The Control Center backend uses a **dual-path gateway client** architecture:
+
+- **Primary path**: WebSocket client (`gateway.WSClient`) connects to the OpenClaw gateway using WS protocol v3. It handles handshake, initial sync (agents.list + sessions.list RPCs), live event routing (sessions.changed, presence, agent.config), and automatic reconnection with exponential backoff.
+- **Fallback path**: REST poller (`gateway.Client`) polls the gateway `/api/agents` endpoint on an interval. It only activates if the WS client fails to connect within 30 seconds of startup.
+
+## Live Gateway Connection
+
+### Startup Sequence
+1. Both WS client and REST client start concurrently
+2. REST client waits 30s for WS readiness signal (`wsReady` channel)
+3. If WS connects successfully → REST client stands down (logs "using WS — REST poller standing down")
+4. If WS fails within 30s → REST client falls back to polling (logs "WS not ready — falling back to REST polling")
+5. If no WS client configured → REST client polls immediately
+
+### WebSocket Client (Primary)
+- Config: `WS_GATEWAY_URL` (default: `ws://host.docker.internal:18789/`), `OPENCLAW_GATEWAY_TOKEN`
+- Protocol: v3 handshake (challenge → connect → hello-ok)
+- Initial sync: `agents.list` + `sessions.list` RPCs → persist → merge → broadcast `fleet.update`
+- Live events: `sessions.changed`, `presence`, `agent.config`
+- Reconnection: exponential backoff (1s → 2s → 4s → ... → 30s max)
+
+### REST Poller (Fallback)
+- Config: `GATEWAY_URL` (default: `http://host.docker.internal:18789/api/agents`), `GATEWAY_POLL_INTERVAL` (default: 5s)
+- Only used when WS is unavailable
+- Polls the `/api/agents` endpoint and syncs agent status changes
+
+### Wiring
+```
+main.go
+  ├── wsClient = NewWSClient(...)
+  ├── restClient = NewClient(...)
+  ├── wsClient.SetRESTClient(restClient)  // WS notifies REST on ready
+  ├── restClient.SetWSClient(wsClient)    // REST defers to WS
+  ├── go wsClient.Start(ctx)              // primary
+  └── go restClient.Start(ctx)            // fallback (waits for WS)
+```
+
+## Key Design Decisions
+- **Push over poll**: WS is preferred for real-time updates; REST is a safety net
+- **DB first, then SSE**: All event handlers persist to DB before broadcasting
+- **Graceful degradation**: System works without WS; REST provides basic functionality
+- **No hard dependency on REST /api/agents**: If WS is connected, REST endpoint is never called