feat(test): enable native tool calling for GPT-OSS invoice extraction

- Update smartai to v0.13.2 (native tool calling support)
- Update smartagent to v1.5.1 (useNativeToolCalling option)
- Enable think: true for GPT-OSS reasoning mode in Ollama config
- Enable useNativeToolCalling: true in DualAgentOrchestrator
- Simplify driver system message (native tools don't need XML instructions)

Native tool calling uses Ollama's built-in Harmony format parser
instead of requiring XML generation, which is more efficient for GPT-OSS models.

test/test.invoices.nanonets.ts

@@ -30,8 +30,10 @@ const smartAi = new SmartAi({
     baseUrl: OLLAMA_URL,
     model: EXTRACTION_MODEL,
     defaultOptions: {
-      num_ctx: 32768,   // Larger context for long invoices + thinking
-      temperature: 0,   // Deterministic for JSON extraction
+      num_ctx: 65536,    // 64K context for long invoices + reasoning chains
+      temperature: 0,    // Deterministic for JSON extraction
+      repeat_penalty: 1.3, // Penalty to prevent repetition loops
+      think: true,       // Enable thinking mode for GPT-OSS reasoning
     },
     defaultTimeout: 600000,  // 10 minute timeout for large documents
   },
@@ -636,7 +638,7 @@ tap.test('Stage 2: Setup Ollama + GPT-OSS 20B', async () => {
   console.log('  [SmartAgent] Starting SmartAi...');
   await smartAi.start();
 
-  console.log('  [SmartAgent] Creating DualAgentOrchestrator...');
+  console.log('  [SmartAgent] Creating DualAgentOrchestrator with native tool calling...');
   orchestrator = new DualAgentOrchestrator({
     smartAiInstance: smartAi,
     defaultProvider: 'ollama',
@@ -652,18 +654,11 @@ tap.test('Stage 2: Setup Ollama + GPT-OSS 20B', async () => {
 CRITICAL RULES:
 1. Output valid JSON with the exact format requested
 2. If you cannot find a value, use empty string "" or 0 for numbers
-3. IMPORTANT: Before completing, validate your JSON using the json.validate tool:
-
-<tool_call>
-  <tool>json</tool>
-  <action>validate</action>
-  <params>{"jsonString": "YOUR_JSON", "requiredFields": ["invoice_number", "invoice_date", "vendor_name", "currency", "net_amount", "vat_amount", "total_amount"]}</params>
-</tool_call>
-
-4. Only complete after validation passes
-
-When done, wrap your JSON in <task_complete></task_complete> tags.`,
+3. Before completing, validate your JSON using the json_validate tool
+4. Only complete after validation passes`,
     maxIterations: 5,
+    // Enable native tool calling for GPT-OSS (uses Harmony format instead of XML)
+    useNativeToolCalling: true,
     // Enable streaming for real-time progress visibility
     onToken: (token, source) => {
       if (source === 'driver') {