feat: TensorRT 固定 batch=4 重构

- tensorrt_engine.py 工业级 Buffer Pool
- preprocessor.py 添加 pad_to_batch4()
- postprocessor.py 支持批量输出
- settings.py 固定 batch_size=4

core/postprocessor.py

@@ -597,20 +597,21 @@ class PostProcessor:
         
         output = outputs[0]
         
-        if len(output.shape) == 3:
+        if output.ndim == 3:
             output = output[0]
         
-        num_detections = output.shape[0]
+        if output.ndim == 2:
+            output = output.reshape(-1)
+        
+        num_detections = output.shape[0] // 85
         
         boxes = []
         scores = []
         class_ids = []
         
         for i in range(num_detections):
-            detection = output[i]
-            
-            if len(detection) < 6:
-                continue
+            start_idx = i * 85
+            detection = output[start_idx:start_idx + 85]
             
             x_center = detection[0]
             y_center = detection[1]
@@ -637,16 +638,16 @@ class PostProcessor:
             y2 = y_center + height / 2
             
             boxes.append([x1, y1, x2, y2])
-            scores.append(total_conf)
-            class_ids.append(class_id)
+            scores.append(float(total_conf))
+            class_ids.append(int(class_id))
         
         if not boxes:
             return np.array([]), np.array([]), np.array([])
         
         return (
-            np.array(boxes),
-            np.array(scores),
-            np.array(class_ids)
+            np.array(boxes, dtype=np.float32),
+            np.array(scores, dtype=np.float32),
+            np.array(class_ids, dtype=np.int32)
         )
     
     def filter_by_roi(