fix.factors.prediction should not be set for the surrogate model learner

[mb706]

... in checkLearner. With fix.factors.prediction set to TRUE, factors that have not been seen before are treated the same as factors that have their requirement not fulfilled. Instead, convertOptPathToDf should make sure that all possible values of a factor are present as levels in the generated data frame. (Which is arguably as.data.frame.OptPathDF's job.) As long as this is not fixed: checkStuff checks that the surrogate model learner has the missings property when the par.set of the function has requirements. However, since fix.factors.prediction is set to TRUE, missings can also occur when the initial design did not cover all possible factor values for all columns. checkStuff therefore currently underestimates how necessary the missings property is. Bonus bug: discretes.as.factor is not an argument of as.data.frame.OptPathDF.

[jakob-r]

One thing after another. So when we have fixed as.data.frame.OptPathDF what is left? And yeah. Somehow discretes.as.factor has been obliberated without notice. But now it's TRUE per default so it's okay and the part can be deleted (after discussion with @berndbischl why this option was deleted)

[jakob-r]

Okay. It was deleted on purpose in https://github.com/berndbischl/ParamHelpers/commit/6aed4565051adffd23751a236ae572a265f7d8fb So I will delete it here as well.