hashcache fix landed

[Martin Pool]

The hashcache-invalidation fix discussed previously has now landed in
PQM. 

I am not sure why this suddenly started failing so often.  Perhaps the
kernel was upgraded on the pqm server, or perhaps the time() went into a
domain where its float representation was rounded differently?

Robert, do you think we could get the version of bzr that drives pqm
upgraded to help with this?

-- 
Martin