docs: clarify isolated installation and startup failures

[LauraGPT]

User Impact

Clarifies FunClip's installation entry points and the distinct failures shown in #147. The original screenshot shows a pip TLS certificate failure before FunASR can be installed; the Windows follow-up already enters FunASR but cuts off before the final exception. Neither supports treating every traceback as another missing package.

• English/Chinese READMEs now use a virtual environment and python -m pip, with an import/argument-parsing check.
• Add a concise bilingual troubleshooting guide for certificates, interpreter mismatches, and failures inside an installed package. Keep certificate verification enabled and redact private package/proxy credentials before sharing diagnostics.
• Explicitly separate FunClip's Transformers 4.x requirements from Nano's native Transformers 5.x environment.
• Correct the stale contributor minimum from 1.3.26 to the existing requirements' 1.4.9 and add documentation checks to PR CI.

Related to #147; do not close it based on this documentation update.

Validation

Signed/DCO head 99fe66b08668ee160d92b2030dbb954391739a1f.

Actual fresh Python 3.12.3 Linux x86-64 environment: seeded the venv with uv, then used python -m pip to install matching torch/torchaudio2.10.0+cpu from the official CPU index followed by unchanged requirements from PyPI. It resolved FunASR1.4.15, Transformers4.57.6, Gradio4.44.1 and huggingface_hub0.36.2. The standard-library venv creation command was also checked independently.

python -m pip check
python -c "from funasr import AutoModel; print('FunASR import OK')"
python funclip/launch.py --help
python -m pytest -q -rs tests --ignore=tests/test_litellm_api.py
git diff --check

Import, help, dependency checks and compilation pass. Application tests: 80 passed, 1 skipped; optional LiteLLM tests were explicitly excluded because that provider is unchanged and its extra is not installed. The 23 related tests also passed. Initial new documentation contracts failed against the old entry points, then passed. GitHub Markdown API rendered the bilingual headings and content; independent review found no P1/P2.

No model inference, microphone, Windows/macOS installation or original-user-environment repair is claimed. Existing deprecation/font diagnostics remain in the evidence. Requirements, runtime code, model weights, releases and production services are unchanged.