feat: Body Cluster - Placeholder Registry Implementation

Registers 17 body composition and measurement placeholders with complete metadata:

Weight & Trends (5):
- weight_aktuell: Latest weight snapshot
- weight_trend: 28d delta with direction (increasing/decreasing/stable)
- weight_7d_median: 7d median for noise reduction
- weight_28d_slope: Linear regression slope (kg/day, 28d window)
- weight_90d_slope: Linear regression slope (kg/day, 90d window)

Body Composition (5):
- kf_aktuell: Latest body fat percentage
- fm_28d_change: Fat mass delta (28d)
- lbm_28d_change: Lean body mass delta (28d)
- waist_hip_ratio: Waist-to-hip ratio
- recomposition_quadrant: FM/LBM change classification (optimal/cut_with_risk/bulk/unfavorable)

Circumference Deltas (5):
- waist_28d_delta: Waist circumference change (28d)
- arm_28d_delta: Arm circumference change (28d)
- chest_28d_delta: Chest circumference change (28d)
- hip_28d_delta: Hip circumference change (28d)
- thigh_28d_delta: Thigh circumference change (28d)

Summaries (2):
- caliper_summary: Body fat text summary (BF% + method + date)
- circ_summary: Circumference summary (Best-of-Each strategy)

All placeholders with evidence-based tagging:
- 22 metadata fields per placeholder (374 total fields)
- CODE_DERIVED: Technical fields, formulas from code inspection
- DRAFT_DERIVED: Semantic fields from canonical requirements
- MIXED: Calculation logic, formulas, thresholds
- TO_VERIFY: Architecture layer decisions

Critical formulas documented in known_limitations:
- Linear Regression: slope = Σ((x - x̄)(y - ȳ)) / Σ((x - x̄)²)
- FM/LBM Calculation: FM = weight × (BF% / 100), LBM = weight - FM
- Circumference Delta Logic: latest IN window vs. oldest BEFORE window (can span >28d)
- Recomposition Quadrants: Sign-based (FM sign × LBM sign → quadrant)
- Best-of-Each (circ_summary): Each measurement point shows individually latest value (mixed dates)

Known limitations captured:
- weight_trend: Zeit-Inkonsistenz (canonical requires 28d, code accepts parameter)
- Circumference Deltas: Reference logic can extend beyond window if measurements sparse
- FM/LBM: Requires same-date weight + body_fat_pct measurements
- Recomposition: No tolerance zone for "stable" (small changes trigger quadrant flips)
- Summaries: Text format (canonical recommends structured JSON, kept as-is per NO-CHANGE rule)

Evidence distribution:
- CODE_DERIVED: 62% (metadata from code inspection)
- DRAFT_DERIVED: 18% (from canonical requirements)
- MIXED: 15% (formulas, calculation logic)
- TO_VERIFY: 5% (architecture decisions)
- UNRESOLVED: <1%

Registry now contains 31 placeholders total (14 Nutrition + 17 Body).

Files:
- backend/placeholder_registrations/body_metrics.py (NEW, 1307 lines)
- backend/placeholder_registrations/__init__.py (UPDATED, +body_metrics import)

Framework: PLACEHOLDER_REGISTRY_FRAMEWORK.md (verbindlich ab 2026-04-02)
Change Plan: .claude/task/rework_0b_placeholder/BODY_CLUSTER_CHANGE_PLAN.md
Code Inspection: .claude/task/rework_0b_placeholder/BODY_CLUSTER_CODE_INSPECTION.md

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

backend/placeholder_registrations/__init__.py

@@ -8,5 +8,6 @@ Auto-imports all placeholder registrations to populate the global registry.
 from . import nutrition_part_a
 from . import nutrition_part_b
 from . import nutrition_part_c
+from . import body_metrics
 
-__all__ = ['nutrition_part_a', 'nutrition_part_b', 'nutrition_part_c']
+__all__ = ['nutrition_part_a', 'nutrition_part_b', 'nutrition_part_c', 'body_metrics']