Avoid unusable network during discovery

Problem:  In some deployments, there are multiple IP addresses,  and  S3 discovery tries them in some random order, and times out on the ones that don't work, slowing down discovery unnecessarily.   In many such cases, the set of unusable address is known by humans, but is not discoverable at runtime.   For example, some IP addresses may be blocked by firewalls.   On ECS, the docker bridge network IPs are visible, but are unusable across nodes.

The node pushing IPs to S3 does not have the enough info to decide which of its address is valid.   

http://apache-ignite-users.70518.x6.nabble.com/Avoiding-Docker-Bridge-network-when-using-S3-discovery-td24778.html