Release FunASR 1.3.16

[LauraGPT]

Summary

• release FunASR 1.3.16 with the client-driven realtime endpoint support merged in Add client-driven realtime endpoints #3245
• package the realtime WebSocket implementation under funasr.bin.realtime_ws
• add the installed funasr-realtime-server command while keeping the existing example path as a compatibility wrapper
• declare the service's direct regex and websockets>=10.4 runtime dependencies
• update the English and Chinese release notes and realtime documentation

Validation

• 22 passed across tests/test_realtime_ws_service.py and tests/test_dynamic_streaming_vad.py
• targeted Ruff, py_compile, and git diff --check pass
• clean-source wheel and sdist both pass twine check; both contain the packaged realtime module, CLI metadata, and direct dependencies
• clean Python 3.12 wheel and sdist installs both pass pip check, import as 1.3.16, and expose funasr-realtime-server --help
• real NVIDIA H100 test passes with Torch 2.10.0+cu128, vLLM 0.19.1, websockets 16.1, and Fun-ASR-Nano-2512@272c57b82523
• the installed CLI completed START -> PCM -> COMMIT -> final, returned 2 partials and a non-empty final for 5.546 seconds of audio, then returned a non-empty final for a 900 ms voiced utterance on the same connection
• the second utterance started at the first utterance's exact 5546 ms endpoint; server-side commit processing was 102.6 ms and 44.9 ms respectively

Candidate artifacts

• wheel SHA256: 4a20e367b962c8ef884498e3813666d15d3823afb3f64f00fdd827563554aa23
• sdist SHA256: 2bf75cb41fa15b96db625dcb0586fdfc9d29235366c8e6a8ff82782a938e3d62

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.