add a seed argument to some adjustments

Working with adjust_probability_calibration() and I get slightly different calibration results across tuning parameters:

# A tibble: 9 × 7
  threshold .metric     .estimator   mean     n std_err .config        
      <dbl> <chr>       <chr>       <dbl> <int>   <dbl> <chr>          
1       0   roc_auc     binary     0.712     10 0.0216  pre0_mod0_post1
2       0   sensitivity binary     1         10 0       pre0_mod0_post1
3       0   specificity binary     0         10 0       pre0_mod0_post1
4       0.5 roc_auc     binary     0.710     10 0.0212  pre0_mod0_post2
5       0.5 sensitivity binary     0.195     10 0.0213  pre0_mod0_post2
6       0.5 specificity binary     0.969     10 0.00612 pre0_mod0_post2
7       1   roc_auc     binary     0.710     10 0.0229  pre0_mod0_post3
8       1   sensitivity binary     0.0248    10 0.0184  pre0_mod0_post3
9       1   specificity binary     0.996     10 0.00297 pre0_mod0_post3

This isn't a big deal, but it could confuse users.

I suggest adding an argument seed = sample.int(10^4, 1) for adjustments that use random numbers. This will be evaluated when the tailor is made and using withr we can fix the stream when the adjustment is trained.