I like the idea of being able to allow specific CSS classes for embedded posts.
Another possible approach for getting an accurate representation of content from external sites on Discourse would be to generate an embed version of each post on the external site. That version would be intended to be displayed in an iframe - similar to what Youtube does by creating a version of each video at https://www.youtube.com/embed/video_id.
Instead of posting HTML to Discourse, or having Discourse attempt to scrape the post’s HTML, the post content from the external site’s embed URL could be displayed in an iframe.