Prioritize whole words in inline string diffs

[jamilahmadzai]

Fixes #29

Summary

• Tokenize inline string diffs into Unicode-aware word, whitespace, and punctuation atoms before applying fast-diff, keeping short changed words intact.
• Refine one-for-one replacements over 30 Unicode code points at character level, preserving useful detail for the long-word exception described in the issue.
• Encode tokens with single UTF-16 code-unit markers so the diff cannot split surrogate pairs, with a character-diff fallback if the marker space is exhausted.
• Cover short words, long identifiers, non-ASCII words, punctuation, whitespace, multiline output, and inverted/container snapshots.

Validation

• npx ava test/diff.js — 58 passed on Node.js 24
• npx --yes --package=node@16 -- node node_modules/ava/cli.js test/diff.js — 58 passed
• npx --yes --package=node@22 --call "node --version && npm test" — lint passed with 4 existing TODO warnings; all 297 tests passed
• git diff --check

The repository lint dependency uses the deprecated util.isDate() API, which was removed in Node.js 24. The full gate was therefore run under Node.js 22, a runtime where that legacy dependency still works.

---

IssueHunt Summary

Referenced issues

This pull request has been submitted to:

• #29: Inline diffing should prioritize diffing whole words

---