The document size limit applies to each document in a multi-document distributed transaction (MongoDB 4.2+), not the overall size of the transaction.
If individual documents you want pre-images for are likely to approach 8MB or more, I would review the reason for document growth and reconsider your data modelling approach. Large documents are often due to anti-patterns like massive arrays or bloated documents.
Alternatively you could avoid using both post-images and pre-images for a collection which has large documents and consider:
Requesting only post-images
Requesting only pre-images in the change stream output and fetching the current document