Correlated-residual posterior predictive draw: Z_rep ~ MVN(mu, Omega), Y_rep = I(Z_rep > 0)

Reuses the state's already-stored Cholesky factor L_omega_s directly (one Cholesky per state, per Part 20's performance requirement) via a single D x N_events matrix multiply, vectorised across every event in that state – no per-event loop.

Usage

.ppc_generate_correlated(mu_s, L_omega_s)