Add Vulkan GPU context probe (Issue GPU-01 #9) behind CTHREADS_GPU.

[K-T0BIAS]

Summary

Closes GPU Issue 1: optional Vulkan context in _ext so we can detect a compute device without linking vulkan-1 (loader is LoadLibrary / dlopen at runtime). Default builds stay CPU-only (CTHREADS_GPU=OFF).

CMake / build

• option(CTHREADS_GPU OFF); when ON: find_package(Vulkan), compile GPU sources, CTHREADS_WITH_GPU=1, Vulkan headers only (no Vulkan::Vulkan link)
• Enable via CMAKE_ARGS=-DCTHREADS_GPU=ON (cmd/PowerShell) or --config-settings=cmake.define.CTHREADS_GPU=ON; wipe build/ when toggling

C++ runtime (src/cthreads/cpp/gpu/)

• Process-wide context: open loader -> vkCreateInstance -> pick compute queue (prefer discrete) -> logical device
• Public API: init / shutdown / available / device_name (mutex + double-checked init; failed init cleans up)

Bindings

• gpu_module.hpp / gpu_module.cpp -> cthreads._ext.gpu
• Wired from module.cpp under #ifdef CTHREADS_WITH_GPU

Python

• cthreads.gpu: soft-import _ext.gpu; available / device_name / init / shutdown
• errors.py: CThreadsGPUError + mapped types (VulkanNotBuiltError, VulkanLoaderNotFound, VulkanNoDevice, VulkanInitFailed, …)
• frontend/Gpu/ stubs for later @Gpu work (not functional yet)

Tests

• tests/unit/test_gpu_context.py: error mapping, not-built soft path, fake _ext.gpu; live Vulkan tests skip when not built or no device

Other

• .clangd for Vulkan/project includes (local IntelliSense)
• .gitignore: ignore *.gif demo artifacts

Test plan

• CPU-only: pip install -e .; from cthreads import gpu; gpu.available() is False; import does not crash
• GPU build (VS Dev Cmd / venv): set CMAKE_ARGS=-DCTHREADS_GPU=ON + pip install -e . -v; log shows cthreads GPU: ON (Vulkan)
• gpu.available() / gpu.device_name() match a real GPU (e.g. vs vulkaninfo)
• python -m pytest tests/unit/test_gpu_context.py -v -rs, passes; live tests skip or pass depending on hardware