As part of WT-17940, I noticed that PALI and the btree would disagree on the page count for the shared metadata table.

When we create a new shared metadata root page during a checkpoint, we remove the old one after the checkpoint metadata is written. But we verify at the LSN we wrote the checkpoint metadata, so PALI doesn't "see" that discard, reports the page as live, and gives us one more page than the btree thinks it has.

The workaround introduced in that ticket simply skips the check for the disagg metadata. One ugly alternative I considered was to allow the caller to specify a fudge factor of one, but that seemed kinda gross.