[March-7]

TextMatch(word_boundaries=True) still used substring membership after tokenizing the text, so cat matched bobcat. The same false positives reached the word descriptors and their report counts.

Match literal items with word-character boundaries in the contains/not-contains path. Preserve the existing text normalization, case and lemma pipeline, substring mode, exact/regex modes, and empty-item/list behavior. Multiword, punctuation, Unicode, whitespace and literal regex-character cases are covered.

Fixes #1971.

Validation on Python 3.12.13, pandas 2.2.3 and NumPy 2.5.4:

• Focused descriptor file: 17 failed / 63 passed before the product change, then 80 passed; 6 skipped and 5 deselected in both runs. The 47 new cases include 17 that failed on the base.
• Separate public-API oracle: 15 failed / 45 passed before, then all 60 row assertions passed. The same 10-row CategoryCount report changed from 6 matches to the expected 3.
• Existing compatibility subset: 24 passed, 26 deselected.
• Ruff 0.3.7 lint/format checks and git diff --check passed.

Configured mypy 1.1.1 is blocked by a syntax error in the installed NumPy type stubs; base and patched runs produce the identical error before effective checking. WordNet is absent, so actual lemmatization cases were excluded; no fake lemmatizer was used. The full repository suite and upstream CI were not run.

Prepared with OpenAI Codex assistance. The reproduction and checks above were run locally.