feat(M4-013,M5-001,M5-002,M5-003): routing e2e tests, agent config loading, model+agent switching

- M4-013: Add routing-e2e.test.ts with 9 integration tests covering the full
  classify → match rules → routing decision pipeline; includes coding→Opus,
  summarization→GLM-5, conversation→Sonnet, cheap-tier→Haiku, /model bypass,
  unhealthy-provider fallback, and research→Codex scenarios

- M5-001: Store resolvedAgentName during session creation when agentConfigId
  is provided; expose agentName on AgentSession and SessionInfoDto; emit
  agentName in session:info from chat.gateway.ts (message handler and
  set:thinking handler); preserve userId and conversationHistory in merged
  options so they are not lost when agent config is applied

- M5-002: Add AgentService.updateSessionModel() to update live session
  modelId metadata; wire it into ChatGateway.setModelOverride() so the
  /model command immediately reflects in session:info; add
  ChatGateway.broadcastSessionInfo() to push updated session:info to all
  clients watching a conversation on model or agent switch

- M5-003: Implement /agent <name> command end-to-end: inject Brain into
  CommandExecutorService; replace stub handleAgent() with real
  brain.agents.findByName() + findById() lookup; call
  agentService.applyAgentConfig() to update live session; emit session:info
  via chatGateway.broadcastSessionInfo(); update tests to mock brain and
  agentService.applyAgentConfig; add AgentService.applyAgentConfig() method

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

apps/gateway/src/agent/agent.service.ts

@@ -352,6 +352,11 @@ export class AgentService implements OnModuleDestroy {
 
     this.sessions.set(sessionId, session);
     this.logger.log(`Agent session ${sessionId} ready (${providerName}/${modelId})`);
+    if (resolvedAgentName) {
+      this.logger.log(
+        `Agent session ${sessionId} using agent config "${resolvedAgentName}" (M5-001)`,
+      );
+    }
 
     return session;
   }