sanlock: fix aborted command clearing
    
    When clearing the remainder of an aborted
    command, non-blocking recv is used to avoid
    blocking the main thread, but this means it
    needs to do a minimal amount of retrying when
    EAGAIN is returned.  It wasn't retrying at all,
    so the remainder of the message was handled as
    a new command, which caused errors when the
    fragment was found to have a bad magic number.
    
    Signed-off-by: David Teigland <teigland@redhat.com>