Databricks should resolve this and introduce two options: soft modifiedBefore and hard modifiedBefore (files that are going to be ingored forever). In addition, this is not explained in the documentation, so it is a bit frustrating as it is not intuitive behaviour.