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/fp4/activation_nvfp4_quant_fusion_kernels.cu

@@ -14,15 +14,10 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
 
-#include <cuda_runtime_api.h>
-#include <cuda_runtime.h>
+ #include "../../vendors/functions.h"
 
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
-
-#include <cuda_fp8.h>
+ 
 #include "dispatch_utils.h"
 
 #include "cuda_utils.h"

csrc/quantization/fp4/nvfp4_blockwise_moe_kernel.cu

@@ -19,9 +19,9 @@
 #include <torch/all.h>
 #include <cutlass/arch/arch.h>
 
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
-#include <c10/cuda/CUDAStream.h>
+#include "../../vendors/functions.h"
+
+
 #include "cutlass_extensions/common.hpp"
 
 #include "cute/tensor.hpp"

csrc/quantization/fp4/nvfp4_experts_quant.cu

@@ -14,15 +14,10 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
 
-#include <cuda_runtime_api.h>
-#include <cuda_runtime.h>
+ #include "../../vendors/functions.h"
 
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
-
-#include <cuda_fp8.h>
+ 
 #include "dispatch_utils.h"
 
 #include "nvfp4_utils.cuh"

csrc/quantization/fp4/nvfp4_quant_entry.cu

@@ -14,7 +14,9 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
+
+#include "../../vendors/functions.h"
+
 
 #if (defined(ENABLE_NVFP4_SM100) && ENABLE_NVFP4_SM100) || \
     (defined(ENABLE_NVFP4_SM120) && ENABLE_NVFP4_SM120)

csrc/quantization/fp4/nvfp4_quant_kernels.cu

@@ -14,15 +14,9 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
+#include "../../vendors/functions.h"
 
-#include <cuda_runtime_api.h>
-#include <cuda_runtime.h>
 
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
-
-#include <cuda_fp8.h>
 #include "dispatch_utils.h"
 
 #include "cuda_utils.h"

csrc/quantization/fp4/nvfp4_scaled_mm_entry.cu

@@ -14,8 +14,10 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
-#include <c10/cuda/CUDAGuard.h>
+
+#include "../../vendors/functions.h"
+
+
 #include "cutlass_extensions/common.hpp"
 
 #if defined ENABLE_NVFP4_SM100 && ENABLE_NVFP4_SM100

csrc/quantization/fp4/nvfp4_scaled_mm_kernels.cu

@@ -14,10 +14,7 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
-
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
+#include "../../vendors/functions.h"
 
 #include "cutlass_extensions/common.hpp"
 

csrc/quantization/fp4/nvfp4_scaled_mm_sm120_kernels.cu

@@ -14,10 +14,8 @@
  * limitations under the License.
  */
 
-#include <torch/all.h>
+#include "../../vendors/functions.h"
 
-#include <ATen/cuda/CUDAContext.h>
-#include <c10/cuda/CUDAGuard.h>
 
 #include "cutlass_extensions/common.hpp"
 

csrc/quantization/fp4/nvfp4_utils.cuh

@@ -16,8 +16,8 @@
 
 #pragma once
 
-#include <cuda_runtime.h>
-#include <cuda_fp8.h>
+#include "../../vendors/functions.h"
+
 
 #define ELTS_PER_THREAD 8