Extended from original NEXUS for distributed execution per study.md roadmap.

cd cuda
mkdir build && cd build
cmake .. -DCMAKE_CUDA_ARCHITECTURES=75  # T4 GPU
make -j
./bin/main  # Tests MatMul etc. (set TEST_TARGET_IDX in main.cu)

• CUDA 13
• NCCL (linked)
• Phantom-FHE (bundled)
• NTL, GMP (apt installed)

Implement RNS limb parallelism in Phantom NTT/key-switching, NCCL input-broadcast/output-aggregation.

Run baseline MatMul test to verify.