Don't write checkpoints or error logs by default when calling ModelTrainer.train()

This is consistent with the default behavior of ModelTrainer.train(), that it doesn't write anything to disk by default.

This doesn't affect normal usage of the feature through the CLI, where it is enabled. This is mostly just kinder on the test suite and in princple other "advanced" usage of the API.

This is a follow-up to the issue I mentioned on !36 (merged): !36 (comment 370905)