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

check_engine_output.py

@@ -0,0 +1,53 @@
+"""检查 TensorRT Engine 输出的实际 shape"""
+import sys
+import os
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+import numpy as np
+
+try:
+    import tensorrt as trt
+    import pycuda.driver as cuda
+    import pycuda.autoinit
+    
+    engine_path = "./models/yolo11n.engine"
+    
+    with open(engine_path, "rb") as f:
+        runtime = trt.Runtime(trt.Logger(trt.Logger.WARNING))
+        engine = runtime.deserialize_cuda_engine(f.read())
+    
+    context = engine.create_execution_context()
+    
+    print("=" * 60)
+    print("Engine Binding Information")
+    print("=" * 60)
+    
+    for i in range(engine.num_bindings):
+        name = engine.get_binding_name(i)
+        shape = engine.get_binding_shape(i)
+        dtype = trt.nptype(engine.get_binding_dtype(i))
+        is_input = engine.binding_is_input(i)
+        
+        print(f"\nBinding {i}: {name}")
+        print(f"  Shape: {shape}")
+        print(f"  Dtype: {dtype}")
+        print(f"  Is Input: {is_input}")
+    
+    print("\n" + "=" * 60)
+    
+    input_shape = engine.get_binding_shape(0)
+    output_shape = engine.get_binding_shape(1)
+    
+    print(f"Input shape: {input_shape}")
+    print(f"Output shape: {output_shape}")
+    
+    input_size = np.prod([max(1, s) for s in input_shape])
+    output_size = np.prod([max(1, s) for s in output_shape])
+    
+    print(f"Input size: {input_size}")
+    print(f"Output size: {output_size}")
+    
+except Exception as e:
+    print(f"Error: {e}")
+    import traceback
+    traceback.print_exc()