feat(ugc): ray_backend=hiprt traces on the GPU with HIPRT and Orochi (optional build)

HIPRT (MIT) behind the CMake option DLU_HIPRT (off): its headers come from its
SDK (HIPRT_ROOT, else ROCm's /opt/rocm) and are copied next to the servers; its
library is loaded when first used (hiprtew), as HIP or CUDA are by Orochi
(MIT, fetched pinned by hash; CUDA when its toolkit is found). The trace
kernels (nearest hit skipping the triangle a ray leaves, any hit) are compiled
the first time and kept in cache/hiprt. One GPU context for the process
(hiprt_device picks the GPU); the workers take turns on it. When HIPRT, the
GPU or a scene's upload fails, Embree is used instead, and the UGC server logs
why at start.

For a GPU the rays go in batches (UgcRays::Scene gets batch queries; the CPU
backends answer them a ray at a time):
- hidden faces: with a batch backend the paths are traced side by side, a
  bounce at a time (the path code split into Start, Scatter and Bounce, the one
  by one tracing unchanged); the same paths with the same random numbers, so
  the same triangles are decided (tested with builtin side by side)
- the occlusion bake and the denoised icons' traced occlusion always ask in
  batches (the same rays, the same results)

Its symbols are hidden: the servers export theirs (-rdynamic), and HIPRT's
library, which has an Orochi of its own, would otherwise call ours.

Check: configure with -DDLU_HIPRT=ON on a machine with ROCm (or HIPRT's SDK)
and an AMD RDNA or NVIDIA GPU; UgcServer --make-model x.lxfml out hiprt; the
UGC tests (hits, hidden faces and occlusion against builtin); a build without
it leaves everything as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

dDashboardServer/routes/SettingsCatalog.cpp

@@ -504,6 +504,7 @@ namespace {
 		c.Add(Labels(Choice(UGC, "hsr_method", "How hidden faces are found", "toolbox: LU Toolbox's paths (above). fast: the model rendered from 42 directions around it, removing what shows in none; much faster, but it also removes faces seen only by bounced light (insides seen through openings, recesses). Staff can pick another for one make when making models again.", "toolbox", { "toolbox", "fast" }), { "LU Toolbox (paths)", "Fast (renders from around)" }));
 		c.Add(Unit(Int(UGC, "hsr_fast_resolution", "Fast method's render size", "The size of each of the fast method's 42 renders; bigger keeps smaller visible faces.", "1024", 64, 4096), "pixels"));
 		c.Add(Labels(Choice(UGC, "ray_backend", "Ray tracer", "What traces the rays of the hidden faces' paths and of the occlusion: builtin (the UGC server's own), embree (Intel Embree on the CPU) or hiprt (the GPU, when the server was built with it and has one; else embree). They give the same results but for rounding. Staff can pick another for one make when making models again.", "builtin", { "builtin", "embree", "hiprt" }), { "Built in", "Embree (CPU)", "HIPRT (GPU)" }));
+		c.Add(Int(UGC, "hiprt_device", "GPU for HIPRT", "Which GPU the hiprt ray tracer uses: 0 is the first HIP (AMD) or CUDA (NVIDIA) device.", "0", 0, 16, true));
 		c.Add(Labels(Choice(UGC, "denoise", "Denoise icons", "oidn (when the server was built with Intel Open Image Denoise; else off): a model's icon is drawn from its colors before the occlusion bake, with its occlusion traced per pixel with a few rays and the noise removed by the denoiser, instead of the baked occlusion. The model itself keeps its baked occlusion (a denoiser only works on images).", "off", { "off", "oidn" }), { "Off", "Open Image Denoise" }));
 		c.Add(Int(UGC, "denoise_samples", "Denoised occlusion rays per pixel", "Occlusion rays traced from each pixel of a denoised icon (before it is scaled down, so 16 times as many per icon pixel); fewer is faster and noisier.", "4", 1, 256));
 		c.Add(Bool(UGC, "bake_ao", "Darken hidden corners", "Ambient occlusion baked into the vertex colors.", true));