fix: separate fixture tracing from behavioral closure

[theislampill]

Summary

• accept explicit Boolean behaviorally_decided states for fixture tracing
• add independent fail-closed behavioral_coverage closure dimension
• regenerate unit dispositions, absorption readiness, burndown, and tranche selection

Measured effect

• reopened units: cu-definite-article-assimilation, cu-grapheme-inventory-and-confusables, cu-nunation-support-orthography, cu-orthographic-connectivity-classes, cu-short-vowel-diacritics-and-vocalization-state
• closed units: 5 -> 0
• fully operationalized lessons: 4 -> 0
• real-consumer partial lessons: 49 -> 53
• behavioral decisions across tranche-001 unit obligations: 0; fixture traces remain present and honest

This is the owner-approved G1 doctrine correction. It does not change linguistic facts, certification, or public projection.

Red-first evidence

Before implementation, focused tests failed because:

• True traces were excluded
• _behavioral_coverage_dimension did not exist
• closure ignored behavioral coverage

Verification

• python -m unittest tools.test_curriculum_closure tools.test_select_tranche -v
• python tools/test_l1l6_consumer_truth.py -v
• python tools/validate_curriculum_l1l6.py
• python tools/select_tranche.py --check
• python tools/check_artifact_ergonomics.py
• python tools/test_weak_root_voice_runtime.py -q (66/66)
• python tools/check_regressions.py (ALL REGRESSION CHECKS PASS)
• git diff --check

Board: #163 and #164 will be updated from this regenerated state.