Add accuracy metric to results stats summary

[mccarthy-m-g]

This PR adds an (optional) accuracy metric to the results stats summary of run_eval() (#30).

The main challenge for this one was defining the API, since unlike other error metrics, the user needs to define the absolute and relative error margins for whether a prediction is considered accurate or not. I opted to only calculate accuracy if the user has supplied the absolute and relative error margins (via a new .stats_summ_options argument and corresponding stats_summ_options() function). This lets us avoid defining universally suitable defaults, which is probably difficult to do appropriately.

The API looks like this:

run_eval(..., .stats_summ_options = stats_summ_options(acc_error_abs = 0.5, acc_error_rel = 0.25)

[roninsightrx]

I opted to only calculate accuracy if the user has supplied the absolute and relative error margins

Agree with this approach, otherwise there is too much chance of users just relying on what is default.

[mccarthy-m-g]

One Q for printing: Right now we're printing accuracy with NAs for each type when it isn't being calculated. Do you like that, or should we remove the accuracy column from printing when it isn't calculated?

To hide it, we just need to add a select statement to print.mipdeval_results_stats_summ(), like:

tibble(x = rep(NA, times = 5), y = 1:5) |> select(where(\(.x) all(!is.na(.x))))
#> # A tibble: 5 × 1
#>       y
#>   <int>
#> 1     1
#> 2     2
#> 3     3
#> 4     4
#> 5     5

[roninsightrx]

would include at least one end-to-end test for run_eval() where we do calculate it

[mccarthy-m-g]

Added here: 070a7bb

[roninsightrx]

lgtm!