add discrete method for survreg to return Matdist

[RaphaelS1]

Tests pass locally, first part of #41

@jemus42 this will fix speed and size issues you're seeing in parametric

[bblodfon]

Tests should pass now (expect some lightgbm ones?).

There is still the issue though when the test set is a bit smaller, eg. 10 observations from the rat dataset, there we get slightly different probabilities using discrete = TRUE (new option) vs FALSE (old option). Can easily be replicated by adding predict(task, rows =...) to the new tests. Maybe not an issue after the conversion? I really don't know. @RaphaelS1 what do you think?

[RaphaelS1]

Tests should pass now (expect some lightgbm ones?).

There is still the issue though when the test set is a bit smaller, eg. 10 observations from the rat dataset, there we get slightly different probabilities using discrete = TRUE (new option) vs FALSE (old option). Can easily be replicated by adding predict(task, rows =...) to the new tests. Maybe not an issue after the conversion? I really don't know. @RaphaelS1 what do you think?

You can't do ML on 10 rows, so I don't really care about this edge case

[bblodfon]

@sebffischer any comments from your side? The issue Raphael commented above has to do with sampling from discrete vs continuous distributions so in general is not an issue and we discussed about it.

[sebffischer]

Sorry I am currently sick and will look at it next monday

[sebffischer]

please also update the NEWS.md file

[RaphaelS1]

@sebffischer @bblodfon thanks for reviews and help, can we merge?