feat: Implement LLM streaming support and enhance event handling in review process

backend/app/agents/reviewer.py

@@ -315,6 +315,17 @@ class ReviewerAgent:
                     print(f"   ⚠️ ПРОПУСК: patch пустой или слишком маленький")
                     continue
                 
+                # Callback для LLM streaming
+                async def on_llm_chunk(chunk: str, file: str):
+                    """Handle LLM streaming chunks"""
+                    if self._stream_callback:
+                        await self._stream_callback({
+                            "type": "llm_chunk",
+                            "chunk": chunk,
+                            "file_path": file,
+                            "message": chunk
+                        })
+                
                 # Analyze diff with PR context
                 pr_info = state.get("pr_info", {})
                 comments = await self.analyzer.analyze_diff(
@@ -322,7 +333,8 @@ class ReviewerAgent:
                     diff=patch,
                     language=language,
                     pr_title=pr_info.get("title", ""),
-                    pr_description=pr_info.get("description", "")
+                    pr_description=pr_info.get("description", ""),
+                    on_llm_chunk=on_llm_chunk
                 )
                 
                 print(f"   💬 Получено комментариев: {len(comments)}")

backend/app/agents/tools.py

@@ -99,7 +99,8 @@ class CodeAnalyzer:
         diff: str,
         language: Optional[str] = None,
         pr_title: str = "",
-        pr_description: str = ""
+        pr_description: str = "",
+        on_llm_chunk: Optional[callable] = None
     ) -> List[Dict[str, Any]]:
         """Analyze code diff and return comments"""
         
@@ -154,13 +155,32 @@ class CodeAnalyzer:
         try:
             print(f"\n⏳ Отправка запроса к Ollama ({self.llm.model})...")
             
-            # Создаем chain с LLM и JSON парсером
-            chain = self.llm | self.json_parser
+            # Собираем полный ответ из streaming chunks
+            full_response = ""
+            chunk_count = 0
             
-            # Получаем результат
-            result = await chain.ainvoke(prompt)
+            print(f"\n🤖 STREAMING AI ответ:")
+            print("-" * 80)
             
-            print(f"\n🤖 ОТВЕТ AI (распарсен через JsonOutputParser):")
+            # Используем streaming
+            async for chunk in self.llm.astream(prompt):
+                chunk_count += 1
+                full_response += chunk
+                
+                # Отправляем chunk через callback
+                if on_llm_chunk:
+                    await on_llm_chunk(chunk, file_path)
+                
+                # Показываем в консоли
+                print(chunk, end='', flush=True)
+            
+            print("\n" + "-" * 80)
+            print(f"✅ Получено {chunk_count} chunks, всего {len(full_response)} символов")
+            
+            # Парсим финальный результат
+            result = self.json_parser.parse(full_response)
+            
+            print(f"\n🤖 РАСПАРСЕННЫЙ результат:")
             print("-" * 80)
             print(json.dumps(result, ensure_ascii=False, indent=2)[:500] + "...")
             print("-" * 80)