test: document rule-extraction case, punctuation and Unicode behavior

[Ricardo-M-L]

Why

Rule extraction is an intentionally limited offline fallback. Existing tests cover common Chinese
and English patterns, but users need a clear matrix of case, punctuation and whitespace behavior.

Difficulty: easy. A bounded tests-and-documentation task; no model, network or Neo4j required.

Scope

Add table-driven regression tests around RuleExtractor for leading/trailing whitespace, mixed
English case, sentence-final punctuation, repeated punctuation, and non-ASCII names. First record
the current expected behavior. Document unsupported forms instead of silently expanding the parser.
Discuss behavior changes separately if a genuine bug is uncovered.

Acceptance criteria

• At least one positive and one negative/unsupported example for relevant categories, with exact
  expected triples and case normalization assertions.
• Tests are deterministic and offline; cargo test and all-feature Clippy remain green.
• Add a small supported/unsupported example table in docs/guide.md and link it from the Chinese README.
• Do not describe pattern matching as general NER or model-level semantic understanding.

Start in src/extract.rs and docs/guide.md. Comment before working to avoid duplication.

Contribution instructions.
Project license: MIT; no additional contributor agreement.