v0.2.0

What's New

Performance

• Analytic inverses for AR(1) and CompSymm correlation matrices -- O(m) instead of O(m^3) dense solves
• Analytic log-determinants -- O(1) for AR(1) and CompSymm
• Batched computation for balanced panels -- single matrix multiply across all groups
• Thread pool parallelism -- pass n_jobs=-1 to .fit() to distribute group-level computation across CPU cores

Testing

• 12-scenario R nlme::gls comparison suite -- direct numerical comparison of coefficients, standard errors, sigma, log-likelihood, AIC/BIC, and correlation/variance parameters against R's nlme::gls() output
• Covers: OLS, AR(1) (moderate/high/negative phi, ML/REML), compound symmetry, VarIdent, combined AR(1)+VarIdent, multiple predictors, unbalanced panels, intercept-only models