Reviewing the SDK, the topics ASCII Data in Text Archive Files and Archive File Format are not entirely clear around this point. I've entered a SDK bug with the following context.
The missing point is that the file itself should be saved as ASCII even if it's going to support a different encoding. Generally a file saved in a specific encoding will start with hex characters that communicate the file encoding to encoding-aware editors. For example,
Another way of expressing this detail is: When trying to build UTF8 MSIs through IDTs, the data (including column/table names) inside this IDT file could be UTF8, but the file itself should not be UTF8.
Content credit also belongs to