Align fatbin layout definitions with CUDA

[kevmo314]

Summary

• replace stale reverse-engineered fatbin declarations with a small shared CUDA-compatible wrapper/header definition
• use the shared fatbin definitions from client and server module/library load paths
• add a CTest layout check against CUDA's public __fatBinC_Wrapper_t from fatbinary_section.h

Verification

• cmake -S . -B build
• cmake --build build --target fatbin_layout_test lupine_driver lupine_driver_server -j$(nproc)
• ctest --test-dir build --output-on-failure -R 'h2_test|fatbin_layout_test'