When the connect attempt fails and backs off, we should start the clock
at the last connection attempt, not time at which we queue up the
reconnect job.
Signed-off-by: NTrond Myklebust <trond.myklebust@primarydata.com>