Create custom [.forecast() method

[Bisaloo]

This is a first draft for #816

Todo:

• Add custom head() and tail() methods to avoid unnecessary warnings
• Add setter methods
• Ensure data.table features are conserved in the new method

Open questions / discussion points:

• How to reduce duplication in ensuring we manipulate the right class? Maybe a new unclass argument in clean_forecast()?
• Double check I'm using data.table correctly and not changing in place when not appropriate

[seabbs]

This looks great so far. data.table usage looks sensible though I think this now creates a copy where it didn't before (i.e every time as.data.table is called). I don't think this is a major issue though.

[nikosbosse]

as.data.table() does make a copy by default

[nikosbosse]

I merged new features into main and then thought it would be a good idea to fix the ensuing merge conflicts here. Maybe I should not have done that, apologies for interfering with your code.

This created a test failure due to the fact that the functions for nominal forecasts I implemented modified forecasts objects in a way that triggered your checks.

I made a PR targeting this PR to fix these issues: #892

[seabbs]

This is great. I earnt some neat stuff from this PR. As far as I am concerned, this is good to go but think we should wait for @nikosbosse to review and confirm.

[nikosbosse]

This is great, thanks so much!

f <- as_forecast_quantile(example_quantile)
f <- f |> 
  dplyr::select(-observed)
f

Just noting that we wouldn't get any warnings when people use dplyr. But I assume that is expected/intended, right?

[nikosbosse]

It feels like replacing the as.data.table() calls eventually would be nice, but we can also make this an issue and address it later.

[seabbs]

oo sorry Nikos I read this first as a approval and not a comment and so hit merge....

I agree as should look at the data.table issue separately.

[nikosbosse]

No worries, I think it's fine to merge. @Bisaloo thanks a lot, this was a really cool PR!
Do you think there is something clever we can do with dplyr as well? I remember looking into what they use under the hood for mutate at some point. But maybe we also don't want to get too far down the rabbit hole...