Cope with CACHEDIR suddenly vanishing
    
    The production openQA server suddenly started barfing because
    fedfind's cache directory had entirely disappeared. Functions
    that write to files in the cache directory didn't expect this
    as a possibility and didn't check the directory actually exists
    before trying to write files in it.
    
    It seems this can happen if we're using a temporary cache dir
    (because we can't write to ~/.cache) and the process runs for a
    long time; I think maybe if /tmp starts getting large, systemd
    starts 'cleaning things up' or something like that. It's easy
    enough to cope with, so let's do it.
    
    Signed-off-by: Adam Williamson <awilliam@redhat.com>