docs: note mixed-device support in PyPI 1.4.16

[LauraGPT]

Summary

Update the English and Chinese Python API version notes: the mixed-device placement fix from #3706 is now available in the published PyPI 1.4.16 package. The source-only instruction added in #3714 was accurate before that release, but now sends users through an unnecessary source installation.

The notes provide a pinned installation command, retain the installation-guide links and 1.4.15 boundary, and preserve the source-commit/import-path reproducibility advice. No runtime, dependency, or historical evaluation-version changes.

Verification

• Downloaded the actual PyPI funasr-1.4.16-py3-none-any.whl and verified its published SHA-256: f95943f6a8111349a1b08907363b048a49e9cb9caa1e3afd8c92970d08bf450d.
• Its funasr/auto/auto_model.py Git blob is 61f7ce8b92116c781d6c1a6ecf517fc80ff3b46b, byte-for-byte identical to the fix: honor submodel device placement across inference calls #3706 merged implementation. Package version is 1.4.16; PyPI upload was September 18, 2026.
• Focused Python API, reference, relative-link, and installation-command documentation checks: 56 passed, 1 Sphinx module skipped (Sphinx is unavailable in this test environment).
• Rendered both complete pages using GitHub GFM and checked the release links, installation command, blockquotes, and tables.
• Independent read-only review found no P1/P2 issues. Signed commit with DCO sign-off.

This is an artifact/source and documentation verification, not a fresh environment installation or model inference benchmark. The original M2/121-minute case in #3701 remains unverified; this PR does not close it or claim a performance fix. An initial bounded test invocation timed out without producing a result; the subsequent bounded run completed successfully in 3.21 seconds.

AI assistance was used for this documentation update and review.