Plot labelling for >9 experiments is currently broken because in the character representation of the experiment numbers, 9 comes after 10, 11, etc, which leads to a reordering of the boxes without a corresponding reordering of the labels. Example:

library(mlr3)
library(mlr3learners)
library(mlr3viz)

design = benchmark_grid(
    tasks = list(tsk("sonar"), tsk("iris"), tsk("german_credit")),
    learners = list(lrn("classif.featureless"), lrn("classif.log_reg"), lrn("classif.ranger"), lrn("classif.rpart")),
    resamplings = rsmp("cv", folds = 3)
)
bmr = benchmark(design)

print(bmr$aggregate())
autoplot(bmr)

In the first panel the featureless learner moves to the end and is mislabelled.

This PR fixes the issue by zero-padding the strings generated from the numbers. Note that this is technically a workaround and will break if someone ever tries to show the results of more than 999,999,999 experiments in one plot.