extract response correctly in fitted.blackboost hatvalues (and thus AICs) for GLMs with centering of covariates may have been wrong since version 0.5-0 add paper examples to tests
