Currently we extract key features of the document to be inserted during bucket_catalog::insert, such as the normalized metaField and timeField. We already do this without holding the stripe lock, so it will be straightforward to extract it to a separate preparation method. This will offer us more flexibility to optimize for batched inserts.