chore: prepare FunASR 1.4.0 release

[LauraGPT]

Summary

• Advance the package version from 1.3.30 to 1.4.0, following the release policy that no numeric segment exceeds 30.
• Add the July 31 release entry to all four top-level READMEs. The wording distinguishes the PyPI package fix (vda_model now points users to vad_model) from the legacy WebSocket update shipped in the tagged GitHub source.
• Add a release contract test covering the exact version, three numeric segments capped at 30, and four-language release links.

Type of change

• Bug fix
• Documentation
• Example or demo
• Runtime or deployment
• Benchmark or evaluation
• Model/training change

Validation

• python -m compileall funasr examples tests
• Docs or links checked
• Runtime/deployment command tested

Additional evidence:

• 84 passed across the release contract, AutoModel, WebSocket finalization, install docs, runtime download, and llama.cpp release workflow tests.
• Built funasr-1.4.0-py3-none-any.whl and funasr-1.4.0.tar.gz; both pass twine check and SHA256 verification.
• Installed the final wheel in an isolated environment and verified package version 1.4.0 plus the vda_model error path.

User impact

Users who misspell vad_model now get an immediate actionable error instead of silently losing VAD-dependent segmentation, speaker processing, and sentence_info. The release notes also make the WebSocket finalization behavior discoverable without implying it is contained in the PyPI wheel.

Notes for reviewers

This PR prepares metadata only; PyPI upload and tag creation remain gated on the exact merged main commit on 2026-07-31. The GitHub Release will attach the wheel, sdist, and the current nine llama.cpp runtime archives together.

[gemini-code-assist]

Caution

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