Add Fold Weights for Variable Resample Weighting

[tjburch]

Closes #990

This PR implements support for variable fold weights in hyperparameter tuning. This is useful in cases where folds may have differing numbers of observations, and you want proportional contribution to hyperparameter selection.

The implementation adds two main functions: add_fold_weights() to attach custom weights to rset objects, and calculate_fold_weights() to automatically compute weights proportional to fold sizes. Weights are stored as .fold_weights attributes and should flow through the existing tuning pipeline.

Core changes are in estimate_tune_results() which now detects weights and uses weighted statistics (weighted mean, weighted standard deviation, effective sample size) when aggregating metrics. Implementation should be backwards compatible and non-breaking.

[topepo]

Hey @tjburch. Thanks for the PR.

We're doing a pretty invasive update the this package that will take some time. We'll look at the PR after things are settled there but it might be another 2-3 weeks.

Is this something time-sensitive for you?

[tjburch]

Nope. Just had some bandwidth staying awake on paternity leave. Review at your leisure, let me know if I can assist otherwise.

[topepo]

Just had some bandwidth staying awake on paternity leave

I estimate that 5% of all my work has been while waiting at the bustop or for some sort of practice to end 😄