docs: distinguish native Transformers model downloads

[LauraGPT]

Summary

Make the native Transformers checkpoint discoverable directly in all four language README model download entries, rather than only inside a code example. Label the native -hf artifact as HF / Transformers and retain the original artifact as HF / FunASR. Keep MLT, ModelScope and GGUF links distinct.

All fenced examples, weights/runtime code, capability descriptions and three-item news remain unchanged. This is not native timestamps/diarization/streaming support or a new runtime release.

Verification

• Eight new entry tests failed before editing; final five offline CI suites: 29 passed. Real resampling passes in the existing native environment; a system-Python NumPy/SciPy ABI failure is retained as environment evidence, not hidden by source changes.
• GitHub's actual GFM rendering verified labeled download links across both repositories' eight READMEs. HF API confirms native/original/MLT/GGUF libraries are transformers/funasr/funasr/gguf respectively.
• All eight README code fences are byte-identical. Website build has the same 271 hashes as current production; no deployment needed.
• Independent review found no P1/P2. Before-push backup, verified signed bundle, signature and DCO retained.