If the CN105 connection to the heat pump is down (i.e. unplugged,
wrongly configured, etc), then the ESP device goes into a hard thrash
loop where it is disconnecting and reconnecting to the HA server 2 times
every second, i.e. it smashes the HA server ad-infinitum. This commit
fixes this by ensuring the heat pump attributes are initialised to at
least something non-random. We set the temperatures to NAN so it is
clear they have not been initialised by heat pump, and thus apparent in
logs etc.