docs: clarify SenseVoice word timestamp alignment

[LauraGPT]

Summary

• Explain SenseVoice words/timestamp pairing in both Python SDK guides without indexing by display characters.
• Add identical guarded output helpers and executable contracts for English word units, empty alignment, missing fields and length mismatches.
• Keep the original three CLI examples and model/runtime boundaries unchanged.

Related to QwenAudio/SenseVoice#215; does not close it or claim all special-symbol cases or timestamp accuracy are fixed.

Verification

• New contracts failed before the helper was added; final Python API suite: 10 passed.
• Related documentation: 39 passed; Sphinx: 46 passed; product-site: 511 passed.
• Exact bilingual helper consumed all three saved official FunASR1.4.15 inference outputs: 242/217/16 pairs each; no new inference or accuracy claim.
• Two builds are byte-identical and static validation passes; six EN/ZH320/390/1440 browser cases pass, representative desktop/mobile screenshots inspected.
• Independent source review found no P1/P2. Signed commit and DCO; rollback bundle and source archive retained.
• Initial serial full-site test hit its150s execution timeout and was retained. After terminal process confirmation, the unchanged full suite passed with4 workers in177.51s. No test failure suppressed.