Log: fencer: don't log assertion if unable to create full request reply

Previously, we would log an assertion and a warning if asked to create a reply
to a NULL request. However there is a possible sequence for this to happen:

• Some nodes are up and some down at cluster start-up
• One node is elected DC and schedules fencing of the down nodes
• Fencing is initiated for one of the down nodes
• One of the other down nodes comes up and is elected DC
• The fencing result comes back and all peers (including new DC) are notified
• New DC tries to create a notification for its client (the controller) but doesn't know anything about the initial request

For now, just log a warning and drop the assertion. Longer term, maybe we
should synchronize in-flight request information when a fencer joins the
process group.