cluster-dns-operator corrupts /etc/hosts when fs full

Cause: The node-resolver pods that add the cluster image registry service's name to the /etc/hosts file on the node filesystem did not properly handle errors when creating a temporary file or updating /etc/hosts. Consequence: When the host filesystem was full, the node-resolver pod could inadvertently remove the localhost entries from /etc/hosts. Fix: Error handling was added to the logic for updating /etc/hosts. Result: The node-resolver pods log an error and exit instead of writing incomplete content to /etc/hosts when the host filesystem is full.