refactor: 统一硬件相关头文件引用

将分散在各文件中的CUDA/HIP/MUSA硬件相关头文件引用统一到vendors目录下的对应头文件中，提高代码可维护性。移除重复的头文件引用，优化构建配置。

csrc/quantization/activation_kernels.cu

@@ -1,6 +1,5 @@
-#include <ATen/cuda/CUDAContext.h>
-#include <torch/all.h>
-#include <c10/cuda/CUDAGuard.h>
+
+#include "../vendors/functions.h"
 
 #include <cmath>
 #include "core/math.hpp"
@@ -9,29 +8,8 @@
 
 #include "quantization/w8a8/fp8/common.cuh"
 
-#include <c10/util/Float8_e4m3fn.h>
 
-#ifndef USE_ROCM
-  #include <cuda_bf16.h>
-  #include <cuda_fp16.h>
-  #include <cuda_fp8.h>
-#else
-  #include <hip/hip_bf16.h>
-  #include <hip/hip_fp16.h>
-  #include <hip/hip_fp8.h>
 
-typedef __hip_bfloat162 __nv_bfloat162;
-typedef __hip_bfloat16 __nv_bfloat16;
-typedef __hip_bfloat16_raw __nv_bfloat16_raw;
-  #if defined(HIP_FP8_TYPE_OCP)
-typedef __hip_fp8_e4m3 __nv_fp8_e4m3;
-typedef __hip_fp8x4_e4m3 __nv_fp8x4_e4m3;
-  #else
-// ROCm 6.2 fallback: only *_fnuz types exist
-typedef __hip_fp8_e4m3_fnuz __nv_fp8_e4m3;
-typedef __hip_fp8x4_e4m3_fnuz __nv_fp8x4_e4m3;
-  #endif
-#endif
 
 #include "core/registration.h"
 namespace vllm {