[manager] fix address when binding to 0

[d4l3k]

This fixes the announced address when the manager binds to port 0. Previously we were announcing an address with port 0 which caused a connection error when restoring from checkpoint.

Test plan:

torchft_lighthouse --min_replicas 2 --join_timeout_ms 1000
CUDA_VISIBLE_DEVICES=0 TORCHFT_LIGHTHOUSE=http://localhost:29510 torchrun --master_port 29502 --nnodes 1 --nproc_per_node 1 --max-restarts 10 train_ddp.py
CUDA_VISIBLE_DEVICES=1 TORCHFT_LIGHTHOUSE=http://localhost:29510 torchrun --master_port 29501 --nnodes 1 --nproc_per_node 1 --max-restarts 10 train_ddp.py

[H-Huang]

lgtm!