I use a Python library called pdfrw to take PDFs apart, notably to work with sheet music. I think when you ask it for a page, it returns the whole thing, and the PDF files that get written are still readable PDFs.

I assume that within the file structure are things that enable an internal integrity check, such as maybe checksums or CRC's, which would necessarily be different if the file contents are different.

You can dump the metadata and then restore it, maybe that'll work for you?

qpdf?

pdftk?