Avoid computation of p-values in `add_relative_skill()`

[elray1]

The main source of this request is the following example where we use add_relative_skill to compute a relative absolute error, which generates a bunch of warnings from p-value computations which are not used or saved in the output:

library(scoringutils)
#> scoringutils 2.0.0 introduces major changes. We'd love your feedback!
#> <https://github.com/epiforecasts/scoringutils/issues>. To use the old version,
#> run: `remotes::install_github('epiforecasts/[email protected]')`
#> This message is displayed once per session.
forecast_quantile <- example_quantile |>
  as_forecast_quantile(
    forecast_unit = c(
      "location", "forecast_date", "target_end_date", "target_type", "model", "horizon"
    )
  )
#> ℹ Some rows containing NA values may be removed. This is fine if not
#>   unexpected.

scores <- forecast_quantile |> 
  score(metrics = get_metrics(forecast_quantile, "ae_median"))

scores_w_rel_skill <- scores |>
  add_relative_skill(compare = "model", by = "location", metric = "ae_median")
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with zeroes
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with zeroes
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with zeroes
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with ties
#> Warning in wilcox.test.default(values_x, values_y, paired = TRUE): cannot
#> compute exact p-value with zeroes

colnames(scores_w_rel_skill)
#> [1] "model"                    "location"                
#> [3] "forecast_date"            "target_end_date"         
#> [5] "target_type"              "horizon"                 
#> [7] "ae_median"                "ae_median_relative_skill"

^{Created on 2025-01-03 with reprex v2.1.1}

It would be nice to skip the computation of p-values in a call to add_relative_skill. I think a solution could involve these changes:

• Add support to compare_forecasts for test_type = NULL, in which case no test is performed. To avoid changing inputs/outputs, it could set pval = NA_real_ in that case.
• In add_relative_skill, when calling get_pairwise_comparisons, pass test_type = NULL.

This is related to discussion in #750, but I'm suggesting smaller changes here.