This should be fixed upstream. Recent upstream uses systemd with socket activation. Basically, this means the the socket is created at an very early stage by systemd and then handed to lircd. There is no cleaning of the socket taking place. Together, this means that the scenario described here shouldn't happen.
That said, as upstream I am not convinced that the very idea to start lircd from udev is reasonable.