Gateway high availability (Plan B)

[changbl]

Long term solution is Plan B (only public VIP, and controller reschedules inter-datacenter and Internet flows upon gateway failure to a living gateway)
Problems with Plan A (master + backup gateways via Keepalived for both private and public VIPs)
(1) Break the assumption of 1-to-1 mappings between switch "IP" and "DPID", since master and backup gateways share a same VIP. Make it hard and tricky to implement.
(2) VMAC generation is based on switch DPID. Regular switches can only communicate with the master or the backup gateway, but not both
(3) bakcup gateway can only be started after all regular switches are connected, otherwise the forwarding flows cannot be setup since DPIDs/VMACs of regular switches are unknown to it.