mds mustfix
-- fix cap release size/mtime vs new open race
- - mtime/size update should hold FileLock in writing state until
- things are flushed to the log...?
- - take "wr" reference ala ScatterLock et al?
- rename slave in-memory rollback on failure
- proper handling of cache expire messages during rejoin phase?
-> i think cache expires are fine; the rejoin_ack handler just has to behave if rejoining items go missing