Allow workflows with formulas

[topepo]

Closes #56

The main API change is that a fitted workflow would now be required. That made the changes a little more invasive than I hoped they would be since it affects documentation and testing.

I could make the testing code a little more simple with some changes that would avoid a bunch of expect_type() and expect_s3_class() calls. Now sure how much more you would like me to make more changes.

On the bright side:

library(tidymodels)
library(workboots)

car_subset <- mtcars[, c("mpg", "disp", "wt")]
lm_wflow <- workflow(mpg ~ ., parsnip::linear_reg())
lm_fit <- fit(lm_wflow, car_subset)
new_car <- data.frame(disp = 150.0, wt = 2.5)

set.seed(1)
new_car_pred <-
  predict_boots(
    workflow = lm_fit,
    n = 2000,
    training_data = car_subset,
    new_data = new_car
  ) %>% 
  summarise_predictions()

new_car_pred
#> # A tibble: 1 × 5
#>   rowid .preds               .pred .pred_lower .pred_upper
#>   <int> <list>               <dbl>       <dbl>       <dbl>
#> 1     1 <tibble [2,000 × 2]>  23.9        17.7        29.8

^{Created on 2024-01-29 with reprex v2.0.2}

[topepo]

This "standalone" file is in the next version of workflows and is designed to be added to other packages. It is tested there and the nocov directives prevent it from decreasing code coverage here.

[topepo]

I moved these to a "helper.R" file. Since it starts with "helper" it gets executed prior to each test file.

[topepo]

I moved most of the expect_error() code to expect_snapshot() since it tends to be more robust to unrelated changes to the package.

[topepo]

We could reduce a lot of these by making a "ptype" list which is just a zero-row slice of the data (e.g. x[0,]). That contains the classes of the overall object as well as the columns.

[topepo]

These were causing an error. I, for the life of me, could not find a way to resolve it. I might be able to ask Thomas about it.