Separate page headers into different pages?

[StephenCleary]

Consider: storing page metadata into separate pages, as opposed to part of the same page as data. Then data pages are just data.

[StephenCleary]

This would work if we added hashes to the LPN-FO tree, along with point-updates on change. This has the disadvantage of potentially doubling folio updates (every data page update now has to update a metadata page).

Only data pages would be headerless; all metadata pages would retain the current folio structure. Both used and free data pages would be headerless.

This does complicate validation. Single-pass validation may no longer be possible. Unless we also add hashes to other structures (making data pages updates even more expensive).