Fix AI workflow retry loops and improve fallback handling

lib/server/ai.test.ts

@@ -119,10 +119,12 @@ describe('ai config and runtime', () => {
       system?: string;
       prompt: string;
       temperature: number;
+      maxRetries?: number;
     }) => {
       expect(input.system).toBe('Use concise style');
       expect(input.prompt).toBe('Analyze this filing');
       expect(input.temperature).toBe(0.4);
+      expect(input.maxRetries).toBe(0);
       return { text: '  Generated insight  ' };
     });
 
@@ -216,4 +218,63 @@ describe('ai config and runtime', () => {
     expect(result.model).toBe('qwen3:8b');
     expect(result.text).toContain('AI SDK fallback mode is active');
   });
+
+  it('falls back to local text when report workload fails with insufficient balance', async () => {
+    const warn = mock((_message: string) => {});
+
+    const result = await runAiAnalysis('Analyze this filing', 'Use concise style', {
+      env: {
+        ZHIPU_API_KEY: 'new-key'
+      },
+      warn,
+      createModel: () => ({}),
+      generate: async () => {
+        throw new Error('AI_RetryError: Failed after 3 attempts. Last error: Insufficient balance or no resource package. Please recharge.');
+      }
+    });
+
+    expect(result.provider).toBe('local-fallback');
+    expect(result.model).toBe('glm-4.7-flashx');
+    expect(result.text).toContain('AI SDK fallback mode is active');
+    expect(warn).toHaveBeenCalledTimes(1);
+  });
+
+  it('falls back to local text when report workload cause contains insufficient balance', async () => {
+    const warn = mock((_message: string) => {});
+
+    const result = await runAiAnalysis('Analyze this filing', 'Use concise style', {
+      env: {
+        ZHIPU_API_KEY: 'new-key'
+      },
+      warn,
+      createModel: () => ({}),
+      generate: async () => {
+        const retryError = new Error('AI_RetryError: Failed after 3 attempts.');
+        (retryError as Error & { cause?: unknown }).cause = new Error(
+          'Last error: Insufficient balance or no resource package. Please recharge.'
+        );
+        throw retryError;
+      }
+    });
+
+    expect(result.provider).toBe('local-fallback');
+    expect(result.model).toBe('glm-4.7-flashx');
+    expect(result.text).toContain('AI SDK fallback mode is active');
+    expect(warn).toHaveBeenCalledTimes(1);
+  });
+
+  it('keeps throwing unknown report workload errors', async () => {
+    await expect(
+      runAiAnalysis('Analyze this filing', 'Use concise style', {
+        env: {
+          ZHIPU_API_KEY: 'new-key'
+        },
+        warn: () => {},
+        createModel: () => ({}),
+        generate: async () => {
+          throw new Error('unexpected schema mismatch');
+        }
+      })
+    ).rejects.toThrow('unexpected schema mismatch');
+  });
 });