Added the claim-first attack(model, term) engine and ranked smallest-kill output.
Added row deletion (ranked, greedy, beam, and grouped), uncertainty, covariate, missingness, measurement-error, placebo, specification, and split attack families.
Added support for lm, glm, aov, lmerMod, glmerMod, and coxph perturbation workflows, plus explicit limited support for htest and anova objects that do not retain refittable source data.