Fixed a bug in the stepwise selection functions

[gregorsteiner]

Theresa made me aware of a bug in backward(), see e.g. the following code:

data(sex2)

data_curr = sex2
FU_fit = logistf(formula = as.formula("case ~ age+oc+vic+vicl+vis+dia"), data = data_curr, flic = TRUE, model = TRUE)
backward(FU_fit)
rm(data_curr)

data_curr = data.frame()

f1 = function() {
  data_curr = sex2
  FU_fit = logistf(formula = as.formula("case ~ age+oc+vic+vicl+vis+dia"), data = data_curr, flic = TRUE, model = TRUE)
  backward(FU_fit)
}

f1()

This causes an error in the eval step because the dataset is not found in the parent environment (if it is also in the global environment, there is no problem). However, I think the eval step is not necessary, so in backward() and forward(), I removed the model.frame() and eval() step. This seems to solve the issue, but might have some unintended consequences elsewhere. So, I will do some more testing before merging this.