flood flow in br-tun table22 incorrect

Pre-conditions:
    Neutron （Pike）,firewall_driver=openvswitch， enable l2_population
    related bug: https://bugs.launchpad.net/neutron/+bug/1834979
    network nodeA ,compute node A,compute node B
    RouterA ,tenant network A(VXLAN) connected to routerA and enable DHCP,VM-A created with network A

Reproduction steps:
    1.According to related bug, we create VM-A in compute node A,and make it shutoff,and port still active
    2.Create VM-B with the same tenant network A and scheduled to compute node A
    3.Check flows in compute node A br-tun table22
    4.Create VM-C with the same tenant network A and schedule to compute node B
    5.Still check flows in compute node A br-tun table22
    6.this moment,DHCP in VM-B will fail

Description:
    1.After step 2,we check flows,at this moment,flood flow got some problem,but still work.Flood flow has too much output tunnel port.It has output tunnel port to compute node B whicth it should`t have.Because at this moment,l2_population doesnt work well.Flood flow was downloaded by "_setup_tunnel_flood_flow".
    2.After step 4,we check flows,at this moment,flood flow is incorrect.It will cause lots of problems such as DHCP will not work as usual.In this case,flood flow will only has one output tunnel port to compute node B,it lost output tunnel port to network node A
    3.According to code,we find this bug (https://bugs.launchpad.net/neutron/+bug/1834979) has bad effect to l2_population .
    4.In normal case,VM-A was shutted and port of VM-A should be DOWN ,port of VM-B should be the only first port of tenant network A in compute node A.
    5.But,in this case,VM-A was shutted ,flood flow related was cleared.Update network made port of VM-A reactive,it cause port of VM-B wont be the only first port of tenant network A in compute node A.This confused l2_population and made it not function as usual.