feat: enforce structured JSON output via AI SDK Output.object

- ai-response-schema.ts: Zod schema for { message, suggestions[] }
- ai-chat.controller.ts: Output.object({ schema }) on streamText
  forces the LLM to return valid JSON matching the schema instead
  of free-form prose. Supervisor mode excluded (uses tools, not schema).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

src/ai/ai-response-schema.ts

@@ -0,0 +1,14 @@
+import { z } from 'zod';
+
+export const aiResponseSchema = z.object({
+    message: z.string().describe('Conversational response text for the agent. Plain text, no markdown.'),
+    suggestions: z.array(z.object({
+        id: z.string().describe('Unique suggestion ID like s1, s2'),
+        type: z.enum(['upsell', 'crosssell', 'retention', 'operational']),
+        title: z.string().describe('Short title for the suggestion pill'),
+        script: z.string().describe('2-3 sentence script the agent can read aloud to the caller'),
+        priority: z.enum(['high', 'medium', 'low']),
+    })).describe('0-4 contextual suggestions based on business rules. Include on first response, update on subsequent.'),
+});
+
+export type AiResponse = z.infer<typeof aiResponseSchema>;