Add new measures

[bblodfon]

• Calibration slope from distribution-matching with KM (idea from Avati 2020 to compare models with the slope as a score, in Chapfuwa 2023 there is a Distr-based KM that is used in one axis of the calibration plot instead of the predictions themselves I think, code)?
  • Read tutorial paper
  • Expected Calibration Error (ECE) and MCE (maximum CE), this and this paper have the formulas
  • Simple O/E ratio (observed to expected to outcomes) paper, sup material, mentions ICI, Root mean squared bias, E
• Adapting machine learning techniques to censored time-to-event health record data - section 4.1 has some calibration metric based on KM
• Blanchet's time-dependent C-index from riskRegression (for single time point only, use distr prediction) - same as Uno's AUC(t) for right-censored data (see issue). This measure is Equation 3 from this paper. We force use of single time point here in mlr3proba so that people don't integrate across time points as this can break the 1-1 correspondence between linear predictors or markers used in the riskRegression::Score() function and the time-to-event distribution (nicely explained in Antolini's paper section 2.2.2).