docs: publish verified native Transformers adoption guides

[LauraGPT]

Summary

• Publish authoritative Chinese/English native Transformers guides and an application-oriented blog explaining checkpoint formats, preprocessing, generation, context and acceptance.
• Distinguish the merged implementation from stable releases: verified stable 5.16.1 lacks native Fun-ASR-Nano; pin merged source fc501343edfccdc840eb8594a6cafa8185c2de53 and official -hf revision d93b302ee7fd505e1b3576120fc142fc6f7820e1.
• Link the guide from four concise README news sections (three items each), Model Zoo, deployment matrices, docs catalogue, GitHub Pages export aliases and website model/blog pages.
• Preserve explicit toolkit/-hf/-vllm/GGUF boundaries; this does not add a server adapter, timestamps or diarization.

Verification

• 361 product-site pytest tests with exact collected-node coverage across 306/28/27 shards; 122 Chromium tests, including both languages at mobile390/desktop1440 and screenshot inspection.
• 280 repository documentation tests (three existing deselections), 46 Sphinx-link tests, guide synchronization check; two validated byte-identical builds, 256 files/192 HTML.
• Independent content review: no unresolved findings. Signed commit, DCO, source/test freeze and verified rollback backups.
• Actual merged-source archive wheel build, installed native-source equality, matching torch/torchaudio2.10+cpu and librosa0.11.0, pip check.
• Unmodified documented CPU generation recipe tested on the official short Chinese sample; English single, Chinese keyword and two-record bilingual batch returned nonempty text/EOS before the token cap. Cached official weight/sample SHA and size rechecked.

Limits

Only two distinct short recordings; no CER/WER, concurrency, GPU or capacity claim. Keywords did not enforce a candidate spelling. Processor matrix retains 19 passes / 1 known upstream empty-list IndexError; docs reject empty input and do not claim to fix it. Historical dependency and browser-harness failures are retained as failures, not passing evidence. Documentation/tests only; no PyPI release or model weight change.