Grouped resamples are inconsistently of a group_* class

[mikemahoney218]

After #315 and #316, we'll have four separate grouped resampling functions:

• group_vfold_cv(), the oldest, produces rsplits of class c("group_vfold_split", "rsplit") and rsets of class c("group_vfold_cv", "rset")
• group_mc_cv(), patterned after that, produces rsplits of class c("grouped_mc_split", "rsplit") and rsets of class c("group_mc_cv", "rset")
• group_bootstraps() (Add group_bootstraps() #316) produces rsplits of class c("group_boot_split", "boot_split", "rsplit") and rsets of class c("group_bootstraps", "bootstraps", "rset") (the rsplit inconsistency is my mistake, and is due to working on Add group_initial_split() and group_validation_split() #315 first)
• group_initial_split() (Add group_initial_split() and group_validation_split() #315) produces rsplits of class c("initial_split", "mc_split", "rsplit"), patterned off of initial_time_split()
• group_validation_split() (Add group_initial_split() and group_validation_split() #315) produces rsplits of class c("val_split", "rsplit") and rsets of class c("validation_split", "rset"), patterned off of validation_time_split()

This is a lot of variation, and as Julia mentioned in #315, we should investigate a way to make this be more consistent.

Personally, I like the maximalist position I accidentally wound up with in group_bootstraps(): grouped resamples should subclass the non-grouped version, keeping all of their classes and adding a group_ class on top. I can't think of any case where a method should dispatch on, e.g., vfold_cv() but error on grouped outputs. The closest I can think of is #79, but in that situation you should provide a group_vfold_cv() method as well, not just error.

[juliasilge]

Let's remember to check and update the classes in compat-vctrs.R and compat-vctrs-helpers.R as part of this.

[mikemahoney218]

I'm wondering if we should subclass rset as well, so that all grouped resamples produce an object with classes c("group_rset", "rset"). That would let functions like what's proposed in #79 error on all grouped objects as a class, rather than have it be specific to each object type. Might be too much future-proofing though.

[juliasilge]

Once we get into this, let's see if that is possible and manageable, because doing it all at once is more likely to end up solving our inconsistency challenges, I think. Definitely a lower priority than at least getting us consistent, though.

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.