docs: add RTX 4090 Fun-ASR community benchmark

[LauraGPT]

Summary

• add a bilingual community-validation card for Fun-ASR Nano on one RTX 4090 to the product-site benchmark source
• report the published SeedTTS EN/ZH throughput, latency, quality, memory, and 30-minute soak evidence
• link the primary benchmark, consumer-GPU roadmap, draft integration PR, and raw JSON artifacts
• state explicitly that the results come from a dirty feature worktree and do not establish released SGLang-Omni support or low-latency streaming deltas
• add a generated-output regression test for both language routes

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

• python3 -m pytest web-pages/product-site/tests -q: 46 passed

• product-site build: 21 product pages built

• product-site validator: 92 pages validated

• repeated builds were byte-for-byte identical

• browser suite: 7 passed

• desktop and 390 px mobile screenshots checked for both languages; automated page/section overflow checks passed

• all four GitHub/Gist evidence URLs returned HTTP 200

User impact

Production evaluators get a reproducible consumer-GPU reference for Fun-ASR Nano without mistaking in-progress SGLang-Omni work for a released integration.

Notes for reviewers

The card intentionally lives outside the deployment registry because SGLang-Omni is not yet a shipped FunASR deployment path. It reuses the existing benchmark evidence-card styles and adds no CSS or navigation changes.

[gemini-code-assist]

Caution

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