michaelcahill Please review. Don't merge, but so we can talk. This changes the checkpoint LSN for WT-1700 from the next not-yet-allocated record, to the LSN of the last record. I was mistaken - this change does not fix all the issues in WT-1183 where we cannot run recovery on an incremental backup and then continue.

In order to get that working we need to initialize the new write_start_lsn to the beginning of the last record in the log. Currently we do not walk the last file looking for that although it would be more correct. On initialization I set write_start_lsn to the same potential LSN when creating the new log file. So there is a window where that LSN is pointing to the not-yet-written checkpoint record.

We could add code to do that walk, since it would only be done once at startup.