The autoscale BIG-IP instances will sync configs. You can review the github templates README for more info on clustering. It happens automatically via CFT template deployment.
https://github.com/F5Networks/f5-aws-cloudformation/tree/master/supported/autoscale/waf/via-lb/1nic/existing-stack/payg
Your questions...
Who will assign the request to the LTMs?
-Yes, it will be an ELB in tier 1 doing basic L4 traffic distribution to tier 2 BIG-IP instances.
How does the member server know its return traffic?
-Each BIG-IP instance has its own IP config. There is no sharing of floating IPs. Therefore, each BIG-IP instance will send traffic to the backend server via its own unique self-IP. Return traffic from server to BIG-IP will talk to the AWS cloud hyervisor and route appropriately to the correct BIG-IP instance.