In discussing backup, I realized we did not have a test that would confirm we correctly handle recovering the log when a new table gets created and used after a backup cursor is open (i.e. log records with an unknown file id would appear in the log).