[FEA]: Add support for pathfinder.find_nvidia_header_directory("profiler") (#1162)

Include the profiler header descriptor and install the matching toolkit extra in
the CUDA 12/13 test groups so profiler headers resolve from pip, conda, and CTK layouts.

Made-with: Cursor

Author: rwgk

cuda_pathfinder/cuda/pathfinder/_headers/header_descriptor_catalog.py

@@ -97,6 +97,12 @@ class HeaderDescriptorSpec:
         header_basename="npp.h",
         site_packages_dirs=("nvidia/cu13/include", "nvidia/npp/include"),
     ),
+    HeaderDescriptorSpec(
+        name="profiler",
+        packaged_with="ctk",
+        header_basename="cuda_profiler_api.h",
+        site_packages_dirs=("nvidia/cu13/include", "nvidia/cuda_profiler_api/include"),
+    ),
     HeaderDescriptorSpec(
         name="nvcc",
         packaged_with="ctk",

cuda_pathfinder/pyproject.toml

@@ -19,7 +19,7 @@ test = [
 ]
 # Internal organization of test dependencies.
 cu12 = [
-    "cuda-toolkit[nvcc,cublas,nvrtc,cudart,cufft,curand,cusolver,cusparse,npp,nvfatbin,nvjitlink,nvjpeg,cccl,cupti]==12.*",
+    "cuda-toolkit[nvcc,cublas,nvrtc,cudart,cufft,curand,cusolver,cusparse,npp,nvfatbin,nvjitlink,nvjpeg,cccl,cupti,profiler]==12.*",
     "cuda-toolkit[cufile]==12.*; sys_platform != 'win32'",
     "cutensor-cu12",
     "nvidia-cublasmp-cu12; sys_platform != 'win32'",
@@ -32,7 +32,7 @@ cu12 = [
     "nvidia-nvshmem-cu12; sys_platform != 'win32'",
 ]
 cu13 = [
-    "cuda-toolkit[nvcc,cublas,nvrtc,cudart,cufft,curand,cusolver,cusparse,npp,nvfatbin,nvjitlink,nvjpeg,cccl,cupti,nvvm]==13.*",
+    "cuda-toolkit[nvcc,cublas,nvrtc,cudart,cufft,curand,cusolver,cusparse,npp,nvfatbin,nvjitlink,nvjpeg,cccl,cupti,profiler,nvvm]==13.*",
     "cuda-toolkit[cufile]==13.*; sys_platform != 'win32'",
     "cutensor-cu13",
     "nvidia-cublasmp-cu13; sys_platform != 'win32'",