Teach title claim-function rules

[theclaymethod]

Summary

• teach Unslop to distinguish precise causal titles from personified container/tool titles
• detect headline-only calendar and program-container agency such as “Week 1 ends with a calibrated eval”
• add positive and false-positive cases, including protection for literal statements such as “Week 1 ends on Friday”
• register the new category in the voice pack and shared benchmark

Validation

• python3 evals/run_adversarial.py --only OWNER-07 --only FP-92
• python3 evals/run_adversarial.py --only DOC-04 --only OWNER-07 --only FP-92
• python3 evals/build_shared_benchmark.py --check
• python3 evals/check_taboo_parity.py
• python3 evals/check_pattern_coverage.py
• python3 evals/check_evals_schema.py
• python3 scripts/check_packs.py
• python3 evals/check_commands.py
• python3 evals/check_seeded_docs.py
• python3 evals/check_pairs.py
• python3 evals/check_silhouette.py

The full adversarial suite passes 479 cases with one expected failure and one existing performance-threshold failure. SPAN-03 measures about 2.14 seconds against a 2.0-second threshold on both this branch and unchanged HEAD.