DSpace 6/7 bulkedit MetadataImport unable to remove duplicate metadata in multi-value fields

[alanorth]

Describe the bug
If an item has duplicate metadata in a multi-value field, for example the same author listed twice:

id,dc.contributor.author[en_US]
b88eb8a6-eb41-4f4f-9f39-9600ca6e82fe,Alan Orth||Alan Orth

... it is not possible to remove the duplicate value via the CSV/bulkedit metadata import. If you correct the item's metadata by removing the duplicate value in the CSV and re-uploading it, DSpace does not detect changes:

$ cat /tmp/duplicate-metadata.csv
id,dc.contributor.author[en_US]
b88eb8a6-eb41-4f4f-9f39-9600ca6e82fe,Alan Orth
$ dspace metadata-import -f /tmp/duplicate-metadata.csv
There were no changes detected

I have verified this issue on DSpace 6.3 and 7.1-SNAPSHOT.

Expected behavior
We should be able to remove duplicate metadata via CSV bulkedit. To work around this we currently have to add a new column with the new data, blank out the original column, then move the data back to the old column (and remove the new column).

Related work
This issue was noticed by at least one user on the dspace-tech mailing list in 2015 (see the thread entitled "No changes were detected" when importing metadata via XMLUI).

I guess the problem is in the compare method in dspace-api/src/main/java/org/dspace/app/bulkedit/MetadataImport.java, but I don't understand the code.