bugfix(audit): word-boundary terminology match — fix false positive on substring

[CryptoJones]

TerminologyUsedCheck did a naive substring search:

if term.lower() not in other_text.lower(): flag()

That false-positives whenever a defined term happens to appear inside
a longer word in some other planning file:

term 'lien' matches 'client', 'reliant', 'aliens'
term 'tier' matches 'tiers', 'outlier', 'frontier'
term 'rebate' matches 'rebated'
term 'state' matches 'estate', 'statement', 'statehouse'

When that happens the term gets silently rated as "used elsewhere" and
NEVER flagged — undermining the whole point of the check, which is to
catch dead glossary entries.

Fix: same word-boundary regex with kebab-friendly boundary chars used
by the patterns.py fix:

(?<![\w-]){re.escape(term)}(?![\w-])

Matches multi-word and kebab-case terms as complete tokens; refuses to
match inside a longer identifier.

Test added: define lien in DOMAIN.md; the clean_project fixture's
rendered files mention client everywhere (which contains lien as
a substring). Pre-fix the check silently swallowed the warning;
post-fix lien is correctly flagged as orphaned.

148/148 tests pass; ruff + mypy clean.