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.

packaging/windows/d4r.ini

@@ -0,0 +1,47 @@
+; d4r settings for this game (Windows). This file lives in the game's d4r folder and applies to
+; this game only. Edit it with Notepad, save, and restart the game. Mistakes are reported in
+; d4r\d4r_nvngx.log and by d4r-check.ps1; an invalid value falls back to the default shown here.
+;
+; Engine inference and Ray Reconstruction are not yet integrated in this Windows build.
+
+[DLSS]
+; DLSS model:
+;
+; Model | DLSS | Recommended modes          | Notes
+; ------+------+----------------------------+--------------------------------------
+; K     | 4    | DLAA / Quality / Balanced  | Default; faster
+; M     | 4.5  | Performance                | More demanding
+; E     | 3    | Quality / Balanced         | Legacy CNN; no native Windows layer acceleration
+; L     | 4.5  | Ultra Performance           | Experimental; not Windows-validated
+;
+; Select the quality mode (Quality, Performance, ...) in the game or the OptiScaler menu (Insert).
+; E and L select NGX preset hints only. They do not enable standalone engine paths.
+Model = K
+; NVIDIA's library stamps "DLSS SDK - DO NOT DISTRIBUTE" over the image; d4r turns that off.
+; true shows it.
+ShowWatermark = false
+
+[Interop]
+; Experimental: lets the game continue while DLSS runs on the GPU instead of waiting for it.
+; Requires a game that uses one D3D12 command queue. Off is the tested setting. Whether it is
+; faster depends on the game and has not been measured broadly; turn it off if the game crashes,
+; stutters or shows a wrong image.
+AsyncInterop = false
+
+[Debug]
+; Check every DLSS output for invalid values (NaN/Inf) on the GPU. Costs some speed; use it only
+; when reporting a problem.
+ValidateOutput = false
+; normal: errors and a summary in d4r\d4r_nvngx.log. verbose: every API call and kernel launch
+; (large log; use it only when asked for a bug report).
+Log = normal
+
+[Paths]
+; Folder for the compiled DLSS kernels. auto = %LOCALAPPDATA%\d4r\cache. The first launch fills it
+; and can take several minutes; later launches are quick.
+CacheDir = auto
+
+[Kernels]
+; Experimental RDNA4 only. Requires a matching token-lane FP8 M bundle; startup refuses this
+; setting if d4r/native/<gfx12>/d4r-native-fp8.txt is absent. Default keeps the FP16 path.
+NativeFp8 = false