Reconsider current API for `prepare_probs_distribution_data ()`

[uriahf]

Relevant for:

• Histogram for Calibration Curve
• Performance Data

Possible Outputs:

Naive:
real_positives
real_negatives
real_censored
real_competing

Adjusted to Censoring:
real_positives_adjusted
real_negatives_adjusted
real_competing_adjusted

[uriahf]

Alternative approaches for competing risks:

• Competing: Consider competing event as a non-event (another form of real-negatives).
• Naive / Crude: Ignore Competing Risks - not event and not non-event.
  https://academic.oup.com/biometrics/article/68/2/388/7390713?login=false

Key point:

• Should I consider Composite as a built-in approach? (tend to yes)

• Define the process that will allow all approaches in a unified and usefull manner.
  Key issues:

• One performance data for each approach? (tend to yes)

• How to relate to censoring?

[uriahf]

• How the input that informs about the state-model should look like?

[uriahf]

• create_mids_and_counts_data_from_probs() should produce 1-KM/CIF estimates instead of naive counts in case of adjustment for censoring.

• How to consider counts regarding resource-constraint/ppcr?