Support static CUDA runtime linking

Detailed description of the requested feature

Add a CMake option (e.g. CUDART_STATIC) to support static linking of the CUDA Runtime library (cudart_static).

Currently, the project defaults to dynamic linking. There is a set(CUDA_USE_STATIC_CUDA_RUNTIME OFF), which is effectively a 'no-op' since the project uses a manual find_library call.
This feature is aligned with the H1 2026 Developer Roadmap, under "Platform Support and Ease of Use".

Timeline

Within a day, once maintainers agree to this request.

Target hardware/use case

All current target hardwares (Edge deployments and 'self-contained' C++ applications).

Benefits aligned with the developer roadmap:

• Portability: Eliminates LD_LIBRARY_PATH issues and version mismatches on edge devices.
• Operational Efficiency: Reduces the complexity of Dockerfiles by reducing runtime library dependencies management.
• Footprint: Bundles core dependencies into a predictable binary.

Locate the manual find_library call for the CUDA Runtime and the existing CUDA_USE_STATIC_CUDA_RUNTIME setting in the CMake configuration. Add the requested CUDART_STATIC option and verify configuration and builds for both static and dynamic runtime linking.