Arguments to `targeted::design` are not passed on to fitfun when using `ml_model$estimate`

[benesom]

All tests pass for

  # specials are correctly handled for estimating method with x, y arguments
  m4 <- ml_model$new(formula = y ~ x + offset(log(w)), estimate = glm.fit,
    intercept = TRUE, specials = "offset", family = poisson())
  m4$estimate(ddata_count)
  fit <- glm(y ~ x + offset(log(w)), data = ddata_count, family = poisson())
  expect_equal(coef(fit), coef(m4$fit))

  # it is currently not possible for the estimate method call to pass arguments
  # to targeted::design inside the fitfun
  m4 <- ml_model$new(formula = y ~ x + offset(log(w)), estimate = glm.fit,
    intercept = TRUE, family = poisson())
  m4$estimate(ddata_count, specials = c("offset"))
  expect_true(all(coef(fit) != coef(m4$fit)))

The specials argument is not passed on to targeted::design because of

          private$fitfun <- function(data, ...) {
            xx <- do.call(
              targeted::design,
              c(list(formula = self$formula, data = data), private$des.args)
            )
            args <- private$update_args(self$args, ...)
            args <- c(list(xx$x), args)

Is this something we would like to fix?