[Feature] Implement CUDA/cuSolver Backend for GPU Acceleration

[csparker247]

Description

Implement a MathBackend that offloads linear system solves to NVIDIA GPUs. This is critical for processing high-resolution meshes where CPU-based solvers (Eigen) become the bottleneck.

Tasks

• Create include/OpenABF/Backend/CuSolver.hpp.
• Implement device memory management (H2D/D2H transfers) within the backend context.
• Map spmv to cusparseSpMV and solve to cusolverDn.
• Add CMake detection for the CUDA Toolkit and conditional compilation flags.