Constant LANES 

pub const LANES: usize = _; // 8usize

Batch width the dispatch is tuned for on this target.

Each width is the point where a batch fills the pipeline it runs on:

    AVX-512         16   the square the shuffle network needs
    SHA extension    8   where interleaving reaches the unit's throughput
    crypto extension 4   the widest batch that still fits in registers
    lane loops       8   what the vectorizer fills with a 256-bit register

Past that point more lanes only add register pressure.