pub fn avg_addu(a: u64, b: u64, sew: Vsew, mode: Vxrm) -> u64
Averaging unsigned add: (vs2 + src) >> 1 with rounding per vxrm.
(vs2 + src) >> 1
vxrm
Uses a 1-bit wider intermediate to avoid overflow; no saturation, no vxsat.
vxsat