feat(saas): add model groups for cross-provider failover

Model Groups provide logical model names that map to multiple physical
models across providers, with automatic failover when one provider's
key pool is exhausted.

Backend:
- New model_groups + model_group_members tables with FK constraints
- Full CRUD API (7 endpoints) with admin-only write permissions
- Cache layer: DashMap-backed CachedModelGroup with load_from_db
- Relay integration: ModelResolution enum for Direct/Group routing
- Cross-provider failover: sort_candidates_by_quota + OnceLock cache
- Relay failure path: record failure usage + relay_dequeue (fixes
  queue counter leak that caused connection pool exhaustion)
- add_group_member: validate model_id exists before insert

Frontend:
- saas-relay-client: accept getModel() callback for dynamic model selection
- connectionStore: prefer conversationStore.currentModel over first available

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

desktop/src/lib/saas-relay-client.ts

@@ -44,7 +44,7 @@ interface CloneInfo {
  */
 export function createSaaSRelayGatewayClient(
   _saasUrl: string,
-  relayModel: string,
+  getModel: () => string,
 ): GatewayClient {
   // saasUrl preserved for future direct API routing (currently routed through saasClient singleton)
   void _saasUrl;
@@ -69,7 +69,7 @@ export function createSaaSRelayGatewayClient(
           emoji: t.emoji,
           personality: t.category,
           scenarios: [],
-          model: relayModel,
+          model: getModel(),
           status: 'active',
           templateId: t.id,
         };
@@ -114,7 +114,7 @@ export function createSaaSRelayGatewayClient(
 
     try {
       const body: Record<string, unknown> = {
-        model: relayModel,
+        model: getModel(),
         messages: [{ role: 'user', content: message }],
         stream: true,
       };
@@ -229,7 +229,7 @@ export function createSaaSRelayGatewayClient(
         role: opts.role as string,
         nickname: opts.nickname as string,
         emoji: opts.emoji as string,
-        model: relayModel,
+        model: getModel(),
         status: 'active',
       };
       agents.set(id, clone);

desktop/src/store/connectionStore.ts

@@ -492,12 +492,22 @@ export const useConnectionStore = create<ConnectionStore>((set, get) => {
 
               const kernelClient = getKernelClient();
 
-              // Use first available model (TODO: let user choose preferred model)
-              const relayModel = relayModels[0];
+              // Use first available model as fallback; prefer conversationStore.currentModel if set
+              const fallbackModel = relayModels[0];
+
+              // 优先使用 conversationStore 的 currentModel，如果设置了的话
+              let preferredModel: string | undefined;
+              try {
+                const { useConversationStore } = require('./chat/conversationStore');
+                preferredModel = useConversationStore.getState().currentModel;
+              } catch {
+                // conversationStore 可能尚未初始化
+              }
+              const modelToUse = preferredModel || fallbackModel.id;
 
               kernelClient.setConfig({
                 provider: 'custom',
-                model: relayModel.id,
+                model: modelToUse,
                 apiKey: session.token,
                 baseUrl: `${session.saasUrl}/api/v1/relay`,
                 apiProtocol: 'openai',
@@ -522,14 +532,23 @@ export const useConnectionStore = create<ConnectionStore>((set, get) => {
 
               set({ gatewayVersion: 'saas-relay', connectionState: 'connected' });
               log.debug('Connected via SaaS relay (kernel backend):', {
-                model: relayModel.id,
+                model: modelToUse,
                 baseUrl: `${session.saasUrl}/api/v1/relay`,
               });
             } else {
               // Non-Tauri (browser) — use SaaS relay gateway client for agent listing + chat
               const { createSaaSRelayGatewayClient } = await import('../lib/saas-relay-client');
-              const relayModelId = relayModels[0].id;
-              const relayClient = createSaaSRelayGatewayClient(session.saasUrl, relayModelId);
+              const fallbackModelId = relayModels[0].id;
+              const relayClient = createSaaSRelayGatewayClient(session.saasUrl, () => {
+                // 每次调用时读取 conversationStore 的 currentModel，fallback 到第一个可用模型
+                try {
+                  const { useConversationStore } = require('./chat/conversationStore');
+                  const current = useConversationStore.getState().currentModel;
+                  return current || fallbackModelId;
+                } catch {
+                  return fallbackModelId;
+                }
+              });
 
               set({
                 connectionState: 'connected',
@@ -540,7 +559,7 @@ export const useConnectionStore = create<ConnectionStore>((set, get) => {
               const { initializeStores } = await import('./index');
               initializeStores();
 
-              log.debug('Connected to SaaS relay (browser mode)', { relayModel: relayModelId });
+              log.debug('Connected to SaaS relay (browser mode)', { relayModel: fallbackModelId });
             }
             return;
           }