dedupe script should remove the dataset identifier

[FuhuXia]

For data.json type of harvest source, the harvester should be able to uniquely identifier a dataset, regardless the dataset state is deleted or not. However, the dedupe process simply mark one dataset deleted, therefore we still have two datasets sharing the same identifier. If we can not purge the duplicated from the system, the next best thing is removing the identifier from the package.

Same thing applies to non-datajson harvest types.

Related to #2989.