vectorized_ford_johnson

This project now uses a structure-of-arrays pipeline for pair normalization and pair ordering:

• AVX2 on x86_64
• NEON on arm64
• scalar fallback when SIMD is unavailable

The vector path on Apple Silicon builds the pair blocks directly from the raw input with a NEON deinterleave step, then applies partner-bounded Jacobsthal insertions on the main chain.

make MODE=fast opti
make MODE=mincmp opti

Example

./PmergeMe-fast $(jot -r 3000 1 3000)
./PmergeMe-mincmp --stats $(jot -r 3000 1 3000)

the paper need to be updated