Fix "surprising" readjustment of the number of slices used to perform RSA/outcome mapping

[ConnectedSystems]

The number of conditioning points / slices / bins to use in RSA and outcome mapping ($S$) get automatically (and silently) adjusted to the maximum number of categorical values found in the model spec.

ADRIA.jl/src/analysis/sensitivity.jl

Lines 465 to 468 in 83028fa

	is_cat = (foi_spec.ptype .== "categorical")
	if any(is_cat)
	S = _category_bins(S, foi_spec[is_cat, :])
	end

This has implications both to expected runtime, and validity of the analysis (an increase to the number of conditioning points generally increases the required number of samples).

Instead, the number of slices should only be increased for categorical/integer factors, not for everything.

The result matrix should instead of adjusted with column labels indicating the bin/quantile/conditioning point/position.
Only the columns that are relevant for a given factor should be filled.

This has implications for how the visualisations handle the result matrix so changes to address this needs careful consideration of visualisation processes.

[Rosejoycrocker]

Would you like me to address this @ConnectedSystems ? This is due to my my fixes for categorical variables so sorry about this. Also it would be good to have this working for my sensitivity analysis of mcda thresholds.