Refreshes the latent sociomatrix Z for an ordinal outcome. The
observed ordinal categories in Y impose only rank information on
Z: every latent value tied to category \(c\) must sit above all
latent values of the next-lower category and below all latent values of the
next-higher category. Within that rank-implied window each entry is redrawn
from its dyadic full conditional, a truncated normal whose mean borrows the
transpose partner through the within-dyad correlation rho and whose
variance is \(1-\rho^2\). Missing dyads carry no rank constraint and are
drawn from the untruncated conditional. The diagonal is nuisance and is
refreshed from an independent normal.