release: kern-llm on PyPI, the runtime and both engine plugins in one wheel

[xiaguan]

For 0.3.0: make kern usable from vLLM and SGLang with one pip install.

• One package. pip install kern-llm gives import kern, kern_vllm and kern_sglang, with the vllm.general_plugins and sglang.srt.plugins entry points registered. It replaces the local maturin build plus two setuptools packages. kern is taken on PyPI by an unrelated project, so only the distribution name changes; the import stays kern. The plugins move under crates/kern-py/python.
• Release pipeline (modeled on pegaflow's):
  • a tag also builds an abi3 manylinux 2.28 wheel per arch (x86_64, aarch64) with PyO3/maturin-action;
  • scripts/check_wheel.sh installs it into a clean venv and checks the version against the tag, that both plugins and entry points are present, and that no CUDA library is linked;
  • the wheels go onto the GitHub release (in SHA256SUMS and the provenance attestation), then to PyPI in a separate job by trusted publishing (no token; PyPI's pending publisher for kern-llm names this repo, release.yml and the pypi environment).
• Docs. docs/release.md covers the wheel, the PyPI step, the post-release gate (conc1 in both engines) and why a failed gate means yank + patch bump. README, vllm.md and sglang.md install with pip install kern-llm.

New actions: PyO3/maturin-action (manylinux build container) and pypa/gh-action-pypi-publish (the PyPA uploader).

Checked locally (aarch64, maturin 1.14.1): the wheel builds, check_wheel.sh passes on it, and it fails on a wheel built without the plugin packages and on a wrong version. This PR touches crates/kern-py, so the Release workflow builds and checks both wheels here without publishing.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-15T07:45:39.272960Z	7455614	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 745561449e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Run wheel checks when Rust dependencies change

Expand this path filter to cover crates/kern-runtime/**, crates/kern-manifest/**, the root Cargo.toml, and Cargo.lock. The new wheel directly compiles both Rust crates and inherits workspace configuration, while the regular CI excludes kern-py from default-members; therefore a dependency API or configuration change can pass every PR check but make the first tagged release fail during the maturin build.

Useful? React with 👍 / 👎.