docs: validate native FunASR on vLLM 0.27.1

[LauraGPT]

Summary

• replace the stale vLLM 0.19.1 deployment contract with the native FunASR path in vLLM 0.27.1+cu129
• publish reproducible H100 evidence for multilingual transcription, two-request concurrency, and hotword prompting
• document the community-converted checkpoint boundary and retain the official FunASR split-engine path
• add a registry regression contract for versions, hashes, commands, evidence, and limitations

H100 validation

• vLLM resolved FunASRForConditionalGeneration with max model length 40,960
• Chinese, English, and Japanese requests returned HTTP 200
• English + Japanese concurrent wall time: 1.123 s after warmup
• repeated 开放时间 hotword changed the ambiguous baseline from 开饭时间 to 开放时间
• exact wheel, checkpoint revision, model weight, and audio hashes are pinned in the validation record

Verification

• python -m pytest -q web-pages/product-site/tests (73 passed)
• product build: 27 pages
• output validator: 102 pages
• Playwright: 13 passed across mobile, tablet, desktop, and wide viewports
• targeted vLLM page visual probe: 1440 px and 390 px, no overflow or console errors
• git diff --check