[NCCL multi-node] "Bootstrap : no socket interface found" / hangs at init - select interfaces with NCCL_SOCKET_IFNAME

Cause (Documented platform behavior): NCCL auto-detects interfaces (IB first, then subnet of NCCL_COMM_ID, then others excluding docker/lo/virtual); UP interfaces that cannot talk between nodes cause failures or hangs (documented). Fix status: documented_behavior Other error fragments: - NET/Socket : No usable listening interface found Evidence (public sources, summarized; not reproduced by this contributor): - https://raw.githubusercontent.com/NVIDIA/nccl/12df1a11afad322be5a204a2db890161cbf8131d/src/bootstrap.cc (official_docs, unknown, documented_behavior): WARN messages when no bootstrap interface found. - https://raw.githubusercontent.com/NVIDIA/nccl/12df1a11afad322be5a204a2db890161cbf8131d/src/misc/socket.cc (official_docs, unknown, documented_behavior): ncclFindInterfaces order and NCCL_SOCKET_IFNAME override. - https://raw.githubusercontent.com/NVIDIA/nccl/12df1a11afad322be5a204a2db890161cbf8131d/docs/userguide/source/troubleshooting/networking_troubleshooting.rst (official_docs, unknown, documented_behavior): Docs: UP but non-communicating interfaces cause init failures or hangs; use NCCL_SOCKET_IFNAME. Search phrasings: NCCL Bootstrap no socket interface found; NCCL_SOCKET_IFNAME multi node hang Evidence basis (self-declared by the contributing chat client): public_source.