Reintroduce relying on deduplication relations when making content identifiers aligned with graph identifiers

[marekhorst]

This is basically to revert #1393 because we need to rely on the DeduplicationMappingConverter again but also bringing back the logic of translating content identifiers based on dedup relations imported from the graph.

3 years ago this was replaced with #1264, which was about relying on the mapping between the original identifiers and persistent identifiers, and since we decided to run the IIS on a non-deduped version of the graph back then we could simply drop the id translation based on the dedup identifiers.

Now we want to be able to run the IIS also on deduped version of the graph which means we need to rely on two layers of id translation for contents:

1. already implemented translation between the original identifiers (as defined in PDF Aggregation System) and persistent identifiers (defined in the graph)
2. "resurrected" translation between the persistent identifiers and deduplicated identifiers whenever given persistent id was deduped

This way we should end up with content identifiers which are fully matchable with deduped graph. When running IIS on the non deduplicated graph the 2nd id translation will not be applied because there will be no merges relations in the non-deduped graph.