feat(v0.9.0): unhackable TruthScore rating system for public knowledge

Content-deterministic, cryptographically-bound quality ratings where scores
are functions of content — not popularity. You cannot game the score without
genuinely improving the knowledge. Every rating is bound to exact content
via SHA256; edit one observation and all ratings reset to zero.

Anti-gaming architecture:
- Self-rating blocked (rater_id ≠ publisher_id, enforced at write time)
- Identity set server-side (never from user input)
- One rating per rater per content version (UNIQUE constraint)
- Logarithmic diminishing returns prevent coordinated manipulation
- Burst detection flags anomalous rating patterns automatically
- Content padding dilutes scores — specificity is rewarded, not volume

New MCP tools: rate_public_knowledge, get_knowledge_ratings,
update_verification. Enhanced: search_public_knowledge (+sort_by,
+min_truth_score, +TruthScore in results). Bridge: ratings export/import
with anti-gaming validation on pull. Task Tray: color-coded TruthScore
badges (🟢>0.7, 🟡≥0.4, 🔴<0.4, ⬜ unrated).

Schema: knowledge_ratings + rating_anomalies tables, 6 migrations.

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

Author: RMANOV

db_utils.py

@@ -43,6 +43,18 @@
 SHARE_TYPES = ("entity", "relation", "all")
 ENTITY_ORIGINS = ("local",)  # "shared:{username}" added dynamically
 
+# v0.9.0: Quality rating constants (HARDCODED — not configurable to prevent gaming)
+VERIFICATION_OUTCOMES = ("confirmed", "contradicted", "inconclusive")
+VERIFICATION_WEIGHTS = {"confirmed": 1.0, "inconclusive": 0.5, "contradicted": 0.0}
+
+# Composite score weights (sealed)
+IQ_WEIGHTS = {"specificity": 0.35, "falsifiability": 0.25, "internal_consistency": 0.25, "novelty": 0.15}
+TIER_WEIGHTS = {"iq": 0.40, "verification": 0.35, "cross_validation": 0.25}
+
+# Anomaly detection
+RATING_BURST_THRESHOLD = 5
+RATING_BURST_WINDOW_HOURS = 24
+
 PRIORITY_RANK = {p: i for i, p in enumerate(TASK_PRIORITIES)}
 
 PRIORITY_COLORS = {