feat(domains): complete deep method body migration for all 7 domain classes

MasteryEngine (~515 lines): mastery trend analysis (gain tracking across
diagnoses), misconception severity trend (risk classification: critical/
high/moderate/low), path quality metrics (prerequisite coverage, action
kind distribution), session budget validation, plan execution validation.

QualityEvaluator (~398 lines): weighted composite health scoring (5 metrics
→ 0-100), metric drift detection across snapshot pairs, regression detection
(previously-passing gates now failing), auto-baseline from last snapshot,
domain trend tracking.

TutorRouter (~412 lines): confidence distribution tracking per action kind
and source, rule-engine vs llm-adapter source effectiveness comparison,
evidence binding scoring (full/partial/insufficient), downgrade/fallback
pattern detection, trace quality flagging (high_quality/acceptable/
needs_review/unreliable/unverified).

MemoryPolicyManager (~537 lines): per-layer memory budget enforcement
(capacity + entry size + tag count limits), layer promotion policy
validation (confidence/access count/age thresholds), eviction rate trend
analysis, memory health scoring (fragmentation/staleness/confidence
distribution/eviction pressure 4-axis → 0-100), write pattern analysis.

Each domain class follows the unified validate → delegate → augment →
diagnostics pattern.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: Jacobinwwey