Linux port of Nabla Examples

[raydelto]

Description

Companion to the Nabla Linux port PR (Devsh-Graphics-Programming/Nabla#1101). The examples' common code and a few examples assume Win32 or MSVC. This PR makes the common code build on Linux (Clang 20, X11/XCB), and the tested examples (01, 02, 08, 21, 23) run there. 09 builds and opens a window but waits for mouse input, which the XCB backend doesn't provide yet. The Win32 window-manager and surface selection is preserved. Three shared changes also apply on Windows: the depth-layout initialization, the 09 clamp expression and test 23's loop bound. Windows/MSVC has not been built.

• common/.../SimpleWindowedApplication.hpp: create IWindowManagerXcb on Linux and log an error when there is no X server (DISPLAY unset). Other platforms log "No window manager is implemented for this platform yet!" and return false instead of hitting #error.
• common/.../MonoWindowApplication.hpp and 08_HelloSwapchain: create CSurfaceVulkanXcb on Linux, CSurfaceVulkanWin32 on Windows.
• common/.../CSwapchainFramebuffersAndDepth.hpp: the depth-stencil attachment layout is a depth/stencil pair, so it is now set through its .depth member. Initializing the pair from a single enum doesn't compile with Clang.
• 01_HelloCoreSystemAsset: use CColoredStdoutLoggerANSI when not on Windows.
• 09_GeometryCreator: core::clamp(gcIndex, 0ull, size()-1) mixes unsigned long long with size_t (unsigned long on LP64) and doesn't deduce. Use size_t(0).
• 23_Arithmetic2UnitTest (all platforms): with requiredSubgroupSize, Vulkan caps the workgroup at requiredSubgroupSize * maxComputeWorkgroupSubgroups (VUID-VkPipelineShaderStageCreateInfo-pNext-02756). The loop went up to maxComputeWorkGroupInvocations for every subgroup size. On Intel (64 subgroups max, SIMD8) that made a 1024-invocation pipeline, a validation error and a GPU hang. The loop now stops at min(maxComputeWorkGroupInvocations, subgroupSize*maxComputeWorkgroupSubgroups). NVIDIA (subgroup 32 only) is unaffected.

Testing

These results come from the recorded runs of the merged fork PRs and their reviews, not one combined re-run. Ubuntu 24.04, Clang 20.1.2 + LLD, Release, LunarG SDK 1.4.363.0 with the Khronos validation layer, on an RTX 3050 (NVIDIA 580.178.04) and an Intel UHD TGL GT1 (Mesa 25.2.8), together with the Nabla Linux port branch.

Example	RTX 3050	Intel UHD
01_HelloCoreSystemAsset	pass	pass
02_HelloCompute	pass	pass
21_LRUCacheUnitTest	pass	pass
23_Arithmetic2UnitTest	pass	GPU hang fixed by this PR; the unmodified run then crashes (exit 139) inside Mesa ANV's shader compiler on one pipeline (driver bug, details in the Nabla PR)
08_HelloSwapchain (X11)	renders; resize, minimize, maximize, close; 0 validation messages	same

01 and 21 are CPU/VFS tests, so passing with each ICD selected doesn't show GPU execution. 09_GeometryCreator builds but waits for a mouse, and XCB input is not implemented yet. Not tested: Windows/MSVC, GCC, Wayland, and any example not listed above.

TODO list:

• Build these changes on Windows/MSVC and check that they preserve the intended behaviour. The one intended change is the corrected workgroup cap in 23, for devices whose maxComputeWorkgroupSubgroups limits the workgroup size. The depth-layout initialization and the 09 clamp also change there.
• Merge before the Nabla PR, then bump examples_tests in Nabla to the merge commit and restore the Devsh submodule URL.
• Decide whether to include the Linux port of 70_FLIPFluids here or in a follow-up PR (still under review on the fork).