Incorrect sigterm handler causes false netsplits for docker swarm mode

[Pistro]

I seems that the sigterm handler for docker swarm does not execute the "vmq-admin cluster leave" command. Thus, whenever the swarm size is reduced (or the swarm code is updated and the IP adresses of the nodes changes), VerneMQ does not realize that the swarm nodes which went offline are not coming back, which causes it to act as if there were a netsplit.

If i look through the code of the "start_vernemq" script, it seems that nodes going offline in a docker swarm used to be handled correctly, but then there was some code refactoring, and now I think it only works for Kubernetes clusters.
If believe the bug was introduced in this commit:
da6c608

Thank you for looking at this issue!

[ioolkos]

@Pistro Thanks for reporting. Breaking Swarm compatibility certainly wasn't the idea... will look into it.