Change replica/updatedb (the heart of replica/scan)
to notice remote file system errors, like the network
going down, and bail out.

Previously, if one machine (say ethel) was doing a 
scan of /n/sources/plan9 and then the network connection
between them was lost, ethel's walk of /n/sources/plan9
would not visit whatever files it didn't get to and 
would think they were deleted.  Now it will notice
the network failure and stop.