Fix AI timeout and token limits for thinking model (kimi-k2.5)

- Increase max_tokens from 4096 to 16384 to accommodate reasoning tokens
- Increase timeout from 90s to 180s for thinking model latency
- Add logging for response diagnostics (content length, reasoning, finish reason)
- Better error message when model exhausts tokens on reasoning

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

backend/src/modules/investment-planning/investment-planning.service.ts

@@ -426,9 +426,11 @@ Based on this complete financial picture, provide your investment recommendation
           model,
           messages,
           temperature: 0.3,
-          max_tokens: 4096,
+          // High token limit to accommodate thinking models (e.g. kimi-k2.5)
+          // which use tokens for internal reasoning before generating output
+          max_tokens: 16384,
         }),
-        signal: AbortSignal.timeout(90000), // 90 second timeout
+        signal: AbortSignal.timeout(180000), // 3 minute timeout for thinking models
       });
 
       if (!response.ok) {
@@ -438,10 +440,15 @@ Based on this complete financial picture, provide your investment recommendation
       }
 
       const data = await response.json() as any;
-      const content = data.choices?.[0]?.message?.content;
+      const msg = data.choices?.[0]?.message;
+      // Thinking models (kimi-k2.5) may return content in 'content' or
+      // spend all tokens on 'reasoning_content' with content=null
+      const content = msg?.content || null;
+
+      this.logger.log(`AI response: content=${content ? content.length + ' chars' : 'null'}, reasoning=${msg?.reasoning_content ? 'yes' : 'no'}, finish=${data.choices?.[0]?.finish_reason}`);
 
       if (!content) {
-        throw new Error('Empty response from AI API');
+        throw new Error('AI model returned empty content — it may have exhausted tokens on reasoning. Try a non-thinking model or increase max_tokens.');
       }
 
       // Parse the JSON response — handle potential markdown code fences