Make posterior_predict_latent etc. use function factories

[seabbs]

rather than requiring a function per distribution as we do now. At the same time we should update to use primarycensored vs the current custom functionality.

[athowes]

See #275.

[athowes]

Comment that this is internal to the way the code works. I think the only consequence to users would be that the more general approach might mean we can relax the restriction on which families work to do predictions. In that sense I think if there are other issues that are more like "breaking" we should prioritise those.

[athowes]

We agreed on an initial package release on 2024-11-21 (tomorrow). This is a somewhat large feature which doesn't appear to have work on it currently. I would suggest it might be future work rather than included in the release.

[seabbs]

See issue442. I see this as on the critical release path as its a large feature change (going from limited to all probability distributions) and required for new models (which can share some elements (like posterior prediction)

[seabbs]

I doon't mind making a small release to tick things off but I I think this needs doing before bringing in users