When cloning the collection, the original instance is not used.
createNewCollection(opCtx, readTimestamp, *catalogEntry) --> initFromExisting(..., /*collection=*/nullptr, ...) 
Creating from scratch from the mdb catalog entry ends up calling storageEngine->markIdentInUse() for each index, which only searches drop pending idents.
Under normal circumstances the index is usually not drop pending, so nothing is found and the cloning fails. Upstream this is interpreted as "decided to not clone".

Basically this means cloning only works for clustered collections with no secondary indexes.