Fixed the IPV6 socket issue in encode server of EPD

[hhzguo]

Motivation

When run EPD on node with IPV6 address, the zmq for IPV6 has the following issues:

1 ipv6 address is not handled correctly in get_zmq_socket_host function if host is none
2 encode server doesn't handle IPV6, which causes the communication problems between encode server and prefill server

Modifications

add zmq.IPV6 to socket

Accuracy Tests

lm-eval run result:

Benchmarking and Profiling

The change doesn't impact the performance.
But I post the benchmark result here for ref.
1 encode server, 1 prefill and 1 decode on the same h200 node.

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review Process

1. Ping Merge Oncalls to start the PR flow. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • /tag-run-ci-label, /rerun-failed-ci, /tag-and-rerun-ci
4. After green CI and required approvals, ask Merge Oncalls to merge.

[gemini-code-assist]

Code Review

This pull request introduces IPv6 support for ZMQ communication and refactors network address handling using the NetworkAddress utility. It also adds error handling for socket operations in the encode_server. A critical issue was identified where exceptions in the socket sending logic are caught and logged but not re-raised, which could lead to silent data loss and make debugging difficult.

[ShangmingCai]

Maybe covered in #21236?

[hhzguo]

Maybe covered in #21236?

partially. You may ask the owner of 21236 to merge the change in network.py to his/her and I will close mine. Thank you