Refactor RecipeCommandTest::testExceptionOnRollback() so that it doesn't rely on buggy core behavior

AFAICT #3427118: If config rollback fails validation, it's impossible to tell what caused the rollback in the first place → didn't fix anything, and just is asserting that the same unhelpful error message that occurs during the rollback is actually printed to the CLI?

Isn't the root cause that \Drupal\Core\EventSubscriber\ConfigImportSubscriber::validateModules() observes that the rollback changes core.extension and then "helpfully" observes that core.extension cannot be modified until the filter.format.plain_text text format stops using the media_embed filter, which core/tests/fixtures/recipes/config_rollback_exception/recipe.yml just added to that text format?

Doesn't that then mean that we:

1. either need to make ConfigImportSubscriber::validateModules() smarter
2. or perform the rollback operation in two stages: first revert everything except for core.extension changes, and THEN perform the core.extension rollback, which would prevent triggering this exception in ConfigImportSubscriber::validateModules()?

Yes - it means things in core need to change.

But that's not in scope of the recipe system. (I had raised this same concern with @alexpott.)

#4: not necessarily — the second option in #3 requires no core changes?

Well it is a bug that you can not uninstall filter format providing modules in one step via the config importer. That's nothing to do with recipes and this situation can happen outside of recipes.