Da, video sam, but found it too complex to analyse in the time frame available at that time, and then I forgot about it.
You are one of few people in the world who directly assign a gateway IP address rather than routing-mark using a mangle rule
This part indicates you haven't understood the VLAN handling on bridges really well:
/interface bridge vlan
add bridge=WIFI-br0 tagged=AP1-NET-eth7,AP2-NET-eth8 untagged=WIFI-br0 vlan-ids=""
The configuration section above defines, per vlan or per group of vlans, whether a given port will be a tagged (trunk) or untagged (access) member of that vlan or group; as only one vlan may run tagless on a port, each vlan with at least one access port must have a separate row in that section.
In your case, nothing happens if you forward frames tagged with vlan 20 to the port to which the ubnt for guest network as it will simply ignore them. All (two) ports of the bridge use the tagless frames to deliver the management subnet. So all in all you never need to tag/untag frames as they enter/leave the bridge, neither you need to policy which VLAN will enter through which port, so you don't need vlan-filtering=yes on the bridge, and hence you don't need the section above.
The rest is fine. It could be done using VRF too, where each of the two /interface vlan would be in its own VRF along with its respective WAN and the management would use the "normal" routing. The advantage of VRF is that you need special configuration to
provide leaks between the VRFs whereas with normal routing you need the firewall to
prevent leaks between the subnets, and the fact that you don't need mangle rules which slow things down a little bit or a lot, depending on whether you need fasttracking or not. But as we talk about 3011 here, loading it so much that you could benefit from use of fasttracking by just two LTE uplinks seems unreal to me.