--saveToys does not save POI value used for toy generation in limit tree

[adewit]

Behaviour should be, according to documentation:

If you include [the --saveToys] option, the `limit` TTree in the output will have an entry corresponding to the state of the POI used for the generation of the toy, with the value of **`quantileExpected`** set to **-2**.

But (at least when using --toysFrequentist) it appears to be saving the best-fit value of the POI after fitting to the toy data set.

See also https://cms-talk.web.cern.ch/t/parameter-values-in-toys-with-toysfrequentist-option/15578/2

[nucleosynthesis]

This also seems to be happening without --toysFrequentist. The problem seems to be that (for example), when running with -M MultiDimFit, the branch for r is set (inside the MultiDimFit code ) after the fit to a toy is performed and so while the RooRealVar is correctly set to the value used to generate, this branch isn't.

Instead if you add the option --trackParameter r, the branch trackedParam_r will behave as expected - i.e the quantileExpected -2 branches will be the values used to generate the toy for the POI "r".

We should at least fix the explanation of the option to explain that the user should use the --trackedParameters option if they want to save the values used in the generation but it might be worth seeing if the behaviour can be fixed so that the branches created by MultiDimFit also reflect the same?