v1.8.0

• add bundled CHOLMOD support and use it for the sparse Cholesky factorizations
• add gamma likelihoods with a varying shape modeled by fixed effects ('gamma_varying_shape', 'hurdle_gamma_varying_shape', 'hurdle_regression_gamma_varying_shape')
• add zero-censored shifted gamma likelihoods 'zero_censored_shifted_gamma' and 'zero_censored_shifted_gamma_varying_shape'
• add joint (y, N) and varying-dispersion Tweedie likelihoods: 'tweedie_joint', 'tweedie_varying_dispersion', 'tweedie_joint_varying_dispersion' (and their '_fixed_p' variants)
• add optional restarts of the lbfgs optimizers ('max_num_restarts_lbfgs' and 'cold_restart_lbfgs') which help when the optimizer terminates prematurely, e.g., when the line search fails
• add the optimizer parameter 'reset_modes_line_search_lbfgs' (off by default), with which the mode finding of every trial point of the lbfgs line search starts at the modes of the current parameters
• reset modes of Laplace approximations after unsuccessful line searches of lbfgs and add a warning / debug message when the optimizer terminates for this reason
• use triangular solves instead of explicit Vecchia inverses for prior and posterior samples, predictive variances and latent mean predictions
• replace explicit inverses by supernodal selected inversion and solves in the Laplace Cholesky gradients and the grouped random effects predictive variances
• add experimental relative CG/SLQ stopping rules ('cg_convergence_criterion', 'cg_rel_tol', 'cg_abs_tol', 'cg_multi_rhs_convergence')
• add likelihood = ‘gaussian_heteroscedastic_fixed_and_random’ (only for gp_approx = “vecchia”)
• various bug fixes, code robustness changes, and efficiency improvements