Add Dynamic DeepHit (time-to-event)

[DrShushen]

Description

Closes #22

• Add DynamicDeepHitTimeToEventAnalysis and the infrastructure and models needed for that.
• Add time-to-event evaluation and benchmarking.

Separately:

• Make data loaders subclass DataLoader class.
• Rename some things to ensure consistent naming for the time-to-event concept.
• Remove some unused legacy code.
• Some additional data transformation helper.

Affected Dependencies

scikit-learn added.

How has this been tested?

• See tests, new additions all covered.

Checklist

• I have followed the Contribution Guidelines and Code of Conduct
• I have commented my code following the van der Schaar Lab Styleguide
• I have labelled this PR with the relevant Type labels
• My changes are covered by tests

[DrShushen]

There are still some additional metrics that need to be added as per #9 - will add that in future PR

[bcebere]

Great work! LGTM!

There is a discussion about time-to-event naming: There is dedicated literature for time-to-event regression, which approximates the date of the event. And is different from the current models you have.

Otherwise, survival models approximate the probability of risk(class 1) or survival(class 0) for multiple horizons. In AutoPrognosis I am approximating the probability of risk(same as here), so these models are named risk_estimation models.

So, it should be clear that you don't approximate the date of the event, neither the survival probability. But the risk probability.