Supervisor rescheduling doesn't do the trick since the p2p socket doesn't clean up reliably.
This connects all nodes to a full mesh.
Test Plan: go run github.com/certusone/wormhole/bridge/cmd/guardiand -network wormhole-dev -port 9000 & go run github.com/certusone/wormhole/bridge/cmd/guardiand -network wormhole-dev -bootstrap /ip4/127.0.0.1/udp/9000/quic/p2p/<id> -port 9001