When trying to fix WT-5461 by not returning onpage value if nothing is found in txn_read, I encountered a bunch of test failures in backup tests and test_cursor_random_deleted_partial in test_cursor_random. The cursor random test first load some data, reopen wiredtiger, do a large truncate, and then walk the cursor randomly. The reopen should clear the txnids from the previous run but I noticed some keys still have the old txnids associating with them. I suspect somehow we touched some keys in a page and updated the write_gen of the page but didn't change the cells in the rest of the page. Please correct me if my theory is wrong.