Fix character-limit chunking for overlong tokens

[blue126]

Fixes #6.

Summary

• hard-split a token that exceeds the configured character limit instead of rejecting the whole audiobook
• use the same rule in the frontend chunk-count estimator
• add backend and frontend regression coverage

Validation

• uv run ruff check src tests
• uv run mypy src
• uv run pytest -q (48 passed)
• npm test -- --runInBand (6 passed)
• npm run build

This keeps every Character limit chunk at or below the provider cap, including URLs, OCR runs, and CJK text without whitespace.