Personally, I would in such case ignore SD1078 for the following reason and keep the column in the dataset.
When a reviewer inspects your annotated CRF and sees that you mapped it to your permissible variable, and than doesn’t find that column in your dataset, all alarm bells will ring.
IMO, “plan to collect” (in CRF) automatically means “submit”.
I would also document that all observations have a missing value, either in the Reviewers Guide, or in a “dataset comment” in the CO dataset - or both.