feat: add d4r inference engines and Windows parity tooling

Add native K/M/RR inference engines, HIP kernels, replay and validation
harnesses, and Windows DLSS setup and runtime support. Include GPU
profiling tools, build and packaging scripts, platform integration
patches, and implementation documentation.

scripts/windows/build-native-engine.ps1

@@ -0,0 +1,47 @@
+[CmdletBinding()]
+param(
+    [string]$HipRoot = $env:HIP_PATH,
+    [ValidateSet('gfx1100','gfx1101','gfx1102','gfx1103','gfx1150','gfx1151','gfx1152','gfx1153','gfx1154','gfx1200','gfx1201')]
+    [string]$GpuArch = 'gfx1101',
+    [int]$Parallel = [Math]::Max(1, [Environment]::ProcessorCount - 1)
+)
+$ErrorActionPreference = 'Stop'
+$repo = [IO.Path]::GetFullPath((Join-Path $PSScriptRoot '../..'))
+if (!$HipRoot) { throw 'Pass -HipRoot or set HIP_PATH to the Windows HIP SDK root.' }
+$HipRoot = [IO.Path]::GetFullPath($HipRoot)
+if (!(Test-Path -LiteralPath (Join-Path $HipRoot 'include/hip/hip_runtime_api.h'))) {
+    throw "HIP headers not found under $HipRoot"
+}
+& (Join-Path $PSScriptRoot 'setup-native-engine.ps1')
+if ($LASTEXITCODE) { throw "Native engine dependency setup failed: $LASTEXITCODE" }
+
+$tools = Join-Path $repo '.tools'
+$cmake = Join-Path $tools 'python/cmake/data/bin/cmake.exe'
+$ninja = Join-Path $tools 'python/ninja/data/bin/ninja.exe'
+if (!(Test-Path -LiteralPath $cmake)) { $cmake = Join-Path $tools 'python/bin/cmake.exe' }
+if (!(Test-Path -LiteralPath $ninja)) { $ninja = Join-Path $tools 'python/bin/ninja.exe' }
+if (!(Test-Path -LiteralPath $cmake) -or !(Test-Path -LiteralPath $ninja)) {
+    throw 'Pinned CMake/Ninja missing. Run scripts/windows/setup-windows-tools.ps1 first.'
+}
+$toolchain = Join-Path $tools 'llvm-mingw-20260922-ucrt-x86_64/bin'
+$cCompiler = Join-Path $toolchain 'x86_64-w64-mingw32-clang.exe'
+$cxxCompiler = Join-Path $toolchain 'x86_64-w64-mingw32-clang++.exe'
+if (!(Test-Path -LiteralPath $cCompiler) -or !(Test-Path -LiteralPath $cxxCompiler)) {
+    throw 'Pinned llvm-mingw compiler missing. Run scripts/windows/setup-windows-tools.ps1 first.'
+}
+$oldPath = $env:PATH
+$env:PATH = "$toolchain;$oldPath"
+try {
+    $build = Join-Path $repo "build/native-engine-windows-$GpuArch"
+    $headers = Join-Path $tools 'Vulkan-Headers-v1.4.341'
+    $volk = Join-Path $tools 'volk'
+    & $cmake --fresh -S $repo -B $build -G Ninja "-DCMAKE_MAKE_PROGRAM=$ninja" `
+        "-DCMAKE_C_COMPILER=$cCompiler" "-DCMAKE_CXX_COMPILER=$cxxCompiler" `
+        "-DD4R_HIP_ROOT=$HipRoot" "-DD4R_GPU_ARCH=$GpuArch" `
+        '-DD4R_BUILD_NATIVE_ENGINE=ON' "-DD4R_VULKAN_HEADERS_ROOT=$headers" "-DD4R_VOLK_ROOT=$volk" `
+        "-DD4R_MINHOOK_ROOT=$(Join-Path $repo 'external/MinHook')"
+    if ($LASTEXITCODE) { throw "Native engine configure failed: $LASTEXITCODE" }
+    & $cmake --build $build --target d4r_engine_k d4r_engine_m d4r_engine_rr --parallel $Parallel
+    if ($LASTEXITCODE) { throw "Native engine build failed: $LASTEXITCODE" }
+} finally { $env:PATH = $oldPath }
+Write-Host "Built native runners in ${build}: d4r-k.exe, d4r-m.exe, d4r-rr.exe."