Lessons learned from implementing SIMD-accelerated algorithms in pure Rust