When timestamps lag (e.g., due to replication lag in MongoDB), we have the situation where inserted data cannot be written to data files (it is not "visible" to reconciliation).

In this case, we want page to be able to split normally, so they don't become too large in memory and cause stalls later.

This was partially implemented for row store leaf pages, but not in the case where the leaf page consists entirely of new insert (i.e., there is no previous page image, just a big skip list of inserts).

Change reconciliation to allow pages to split in this case. This will necessarily create empty page images.