Function eval_lanewise_fadd_f16x4 

pub fn eval_lanewise_fadd_f16x4(a: u64, b: u64) -> u64

Packed-float add for F16×4 — the SPEC: four independent scalar binary16 adds. Lanes are isolated (a NaN/Inf/subnormal in one cannot perturb another); the parallel fp16 FPU lanes realize it in silicon.