fix(types): tolerate null end in sentiment analysis results

[mez171]

Summary

The API can return a sentiment_analysis_results item with "end": null (seen on a Universal-3.5 Pro transcript with speaker labels; same root cause as CSAT-447, where a word got 0/0 timestamps). Sentiment.end is typed int, so Transcript.get_by_id fails to parse the entire transcript:

sentiment_analysis_results.0.end
  Input should be a valid integer [type=int_type, input_value=None, input_type=NoneType]

The SDK already works around the same problem for start (CSAT-250) with a before validator on Word that converts null to 0. This PR extends that validator to end, for both pydantic v1 and v2. Field types are unchanged (end stays int), so this is non-breaking for callers.

A matching API-side fix (stop returning null end) is being done separately in the API repo.

Changes

• assemblyai/types.py: Word's null-to-0 validator now covers start and end.
• tests/unit/test_sentiment_analysis.py: add test_sentiment_analysis_null_end, mirroring the existing test_sentiment_analysis_null_start.

Testing

• pytest tests/unit: 739 passed (the new test fails without the fix).
• ruff format --check / ruff check with ruff.toml clean on assemblyai/types.py.

🤖 Generated with Claude Code

https://claude.ai/code/session_013m8gQdRBqzz5kme9fBgo63

---

Generated by Claude Code