[Doc] Update README and docs for v0.28.0

[hsliuustc0106]

Purpose

Prepare the user-facing README and documentation for the stable v0.28.0 release:

• announce v0.28.0 and summarize its verified user-facing highlights
• refresh the README and docs landing-page model examples for the new TTS, diffusion, and robotics coverage
• finalize the pre-built wheel list and the vLLM-Omni 0.28.x ↔ vLLM 0.28.x compatibility text
• link the live rolling and commit-pinned CUDA nightly Docker images built from main
• update the upstream vLLM configuration and environment-variable references to v0.28.0
• align NPU installation examples with the v0.28.0 vLLM-Ascend images and remove the stale v0.26 source-stack instructions
• keep the latest-main NPU path on the already aligned vLLM-Ascend v0.28.0 image instead of documenting a vLLM-Ascend source release branch that does not yet exist

Release coordination

The upstream vLLM v0.28.0 release, documentation, CUDA wheel, and the vLLM-Ascend v0.28.0 A2/A3 CI images were verified. The vLLM-Omni v0.28.0 tag, PyPI wheel, and stable ascend/vllm-omni images are release-produced artifacts and were not live when this PR was prepared; this change should land alongside their publication.

Test Plan

vLLM Version: 0.28.0

vLLM-Omni Commit: 65fab1d14b5fa96579a74955af2c8bd8fd7f76a5

• Run the repository pre-commit gates on all seven changed documentation files.
• Build the full MkDocs site in strict mode from an isolated Python 3.12 environment.
• Inspect the rendered landing, GPU, NPU, configuration, and environment-variable pages.
• Search the installation/configuration docs for stale v0.26 release-line references.
• Verify the upstream vLLM release/docs/wheel and vLLM-Ascend A2/A3 CI image tags.
• Verify the rolling and commit-pinned CUDA nightly tags on Docker Hub.
• Run the repository's full precheck-pr self-review.

Test Result

• Non-Markdown pre-commit gates: passed for all seven changed files.
• Markdownlint: passed for the root README and configuration pages. The remaining changed include/landing files retain existing baseline findings for MkDocs snippet markers, the existing empty anchor, and an existing extra blank line; the release diff adds no new finding.
• Strict MkDocs build for the release-doc changes: passed in 404.60 seconds; the focused GPU-page render after the nightly-link follow-up also passed.
• Render inspection: the landing-page model examples, v0.28 compatibility text, wheel list, NPU commands, and both upstream vLLM links render correctly.
• Stale-reference search: only the intentional v0.26 release-history entry and wheel-history entry remain in the scoped README/docs paths.
• Full precheck: passed with no blockers or warnings; the diff adds no code, Python examples, dependencies, public APIs, runtime state, or simplification candidates.
• Verified live inputs: vLLM v0.28.0 release and CUDA wheel, vLLM v0.28.0 configuration pages, quay.io/atlas-ci/vllm-ascend:v0.28.0{,-a3}, and the multi-architecture vllm/vllm-omni:{nightly,nightly-<commit>} tags. Docker Hub currently has no corresponding ROCm nightly tag.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.