warning for novel levels appear even if you use `step_novel()`

If you use step_novel() in a workflow, you can get the warning Warning: Novel levels found in column 'x': '3'. The levels have been removed, and values have been coerced to 'NA'., which seems like a mistake since step_novel() should be able to take care of it.

I think the warning comes from hardhat::scream().

library(tidymodels)

data <-
  data.frame(
    y = as.factor(c(0, 1, 0, 0, 0)),
    x = as.factor(c(0, 1, 2, 0, 0))
  )

test <-
  data.frame(
    x = as.factor(c(0, 1, 2 ,3))
  )

model_spec <- decision_tree() %>%
  set_engine("rpart") %>%
  set_mode("classification")

rec_spec <- 
  recipe(y ~ x, data = data) %>%
  step_novel(x)

rf_workflow <- 
  workflow() %>%
  add_recipe(rec_spec) %>%
  add_model(model_spec)

final_model <- fit(rf_workflow, data)

test_predict <- predict(final_model, test)
#> Warning: Novel levels found in column 'x': '3'. The levels have been removed,
#> and values have been coerced to 'NA'.

test_predict
#> # A tibble: 4 x 1
#>   .pred_class
#>   <fct>      
#> 1 0          
#> 2 0          
#> 3 0          
#> 4 0

^{Created on 2021-07-04 by the reprex package (v2.0.0)}

Reproduce the warning with the workflow and recipe shown in the issue, then trace the prediction path through hardhat::scream(). Confirm how step_novel() handles the unseen level and whether the warning is still appropriate; done means the example no longer emits a misleading warning while prediction remains unchanged.