This is necessary for long-running processes like hydra-queue-runner:
if a nix-daemon worker is killed, we need to stop reusing that
connection.