Draft of Python API for NVBench

The prototype is based on pybind11 to minimize boiler-plate
code needed to deal with move-only semantics of many nvbench
classes.

python/pyproject.toml

@@ -0,0 +1,58 @@
+[build-system]
+requires = ["scikit-build-core>=0.10", "setuptools_scm"]
+build-backend = "scikit_build_core.build"
+
+[project]
+name = "pynvbench"
+description = "CUDA Kernel Benchmarking Package"
+authors = [{ name = "NVIDIA Corporation" }]
+classifiers = [
+  "Programming Language :: Python :: 3 :: Only",
+  "Environment :: GPU :: NVIDIA CUDA",
+  "License :: OSI Approved :: Apache Software License",
+]
+requires-python = ">=3.9"
+dependencies = [
+  # pathfinder
+  "cuda-bindings",
+
+  # Library expects to find shared libraries
+  # libcupti, libnvperf_target, libnvperf_host
+  # pathfinder is used to find it in the Python layout
+  "nvidia-cuda-cupti-cu12",
+
+  # The shared library
+  # libnvidia-ml must be installed system-wide
+  # (Debian package provider: libnvidia-compute)
+]
+dynamic = ["version"]
+readme = { file = "README.md", content-type = "text/markdown" }
+
+[project.optional-dependencies]
+test = ["pytest", "cupy-cuda12x", "numba"]
+
+[project.urls]
+Homepage = "https://developer.nvidia.com/"
+
+[tool.scikit-build]
+minimum-version = "build-system.requires"
+build-dir = "build/{wheel_tag}"
+
+[tool.scikit-build.cmake]
+version = ">=3.30.4"
+args = []
+build-type = "Release"
+source-dir = "."
+
+[tool.scikit-build.ninja]
+version = ">=1.11"
+make-fallback = true
+
+[tool.scikit-build.metadata.version]
+provider = "scikit_build_core.metadata.setuptools_scm"
+
+[tool.setuptools_scm]
+root = ".."
+
+[tool.scikit-build.wheel.packages]
+"cuda/nvbench" = "cuda/nvbench"