Measures: the whole NA handling seems overly complicated

[berndbischl]

Measures have a property whether they can return an NA, but also a slot to define the value which is returned instead of an NA....

I am not sure this is of ANY use. MAYBE one can have one concept for this in the superclass, to transform an NA or NAN into a different value.

but having separate options for all metrics seems way too much and confusing

[berndbischl]

1. probably this feature might be reasonable to impute NaNs values.

2. the docs of the property say: the measure could return a NaN or NA. we likely only want one. NaN.

3. Might be good to add a sentence in ?Measure that this exists.

4. we need a test for the interaction between this and the NaN production if in Measure$score has weird NaN producing code #1110

[berndbischl]

also rename the propery to nan_value

[berndbischl]

check in mlr3measures that we always produce NaN and not NA