Invent a better way to retain bytes due to limitations of Annotator

[lehins]

Due to applicative nature of Annotator it is impossible to look into the item a that is being decoded if all you have available is the Decoder s (Annotator a). This limitation prevents us from guarding against duplicate detection in Maps and Sets.

Moreover there is real performance overhead of using Annotator, since the way it works is by constructing a huge closure that can later be applied to a ByteString

By recent switch to versioned decoder we have a better way to deal with this issue. A proof of concept has been implemented in this branch: https://github.com/IntersectMBO/cardano-ledger/tree/lehins/alternative-to-annotator