Bundle Punkt models for offline sentence tokenization

[markbackman]

Summary

• Bundle all 19 Punkt models so sentence aggregation works without runtime downloads or an external NLTK data directory. Read parameters directly from the compressed archive without changing NLTK data paths.
• Preserve lazy NLTK imports and background warm-up. Include the pinned archive and provenance in both wheels and source distributions.

Release consideration

NLTK explicitly identifies the Punkt model data license as unclear, separately from its code license: https://github.com/nltk/nltk_data/blob/gh-pages/LICENSE-OVERVIEW.md. Redistribution licensing needs resolution before release; this PR is a draft for review.

Testing

• 33 tests passed covering offline loading, all 19 models, lazy imports, sentence detection, and streaming aggregation.
• Built a wheel from the source distribution and verified tokenization directly from the wheel with external NLTK data disabled and downloads blocked.

Related: #5720.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/pipecat/utils/string.py	90.17% <100.00%> (+2.79%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.