release: v0.7.0 — PPR, Evidence Chain, Auto-Ontology 최적화, HotPotQA 0.856

## v0.7.0 주요 변경사항

### 신규 기능
- Personalized PageRank 엔진 — spreading activation 교체, multi-hop +28%
- Evidence Chain Assembly — BFS bridge + context compression + fact extraction
- HybridClassifier — 규칙(confidence≥0.6) → LLM fallback 2단계 분류
- LLMClassifier batch 모드 — 8~16개 문서 일괄 분류, 비용 3~5x 절감
- EmbeddingRelationDetector — cosine similarity 기반 자동 관계 생성
- PhraseExtractor — HippoRAG2 dual-node KG (passage+phrase)
- AnthropicLLMProvider — Claude API 지원

### 벤치마크 결과
- HotPotQA 24문항 Correctness: 0.856 (Cognee 0.925의 92.5%)
- 4B 로컬 모델(qwen3.5:4b)로 달성
- 외부 데이터셋 8종 IR 벤치마크 포함

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: SonAIengine

pyproject.toml

@@ -4,8 +4,8 @@ build-backend = "hatchling.build"
 
 [project]
 name = "synaptic-memory"
-version = "0.6.0"
-description = "Brain-inspired knowledge graph for LLM agents — auto-ontology (rule/embedding/LLM), Hebbian learning, memory consolidation, hybrid search."
+version = "0.7.0"
+description = "Brain-inspired knowledge graph for LLM agents — PPR, evidence chain, auto-ontology (rule/embedding/LLM), Hebbian learning, memory consolidation."
 license = "MIT"
 requires-python = ">=3.12"
 readme = "README.md"
@@ -15,7 +15,8 @@ keywords = [
     "agent-memory",
     "auto-ontology",
     "hebbian-learning",
-    "spreading-activation",
+    "personalized-pagerank",
+    "evidence-chain",
     "memory-consolidation",
     "hybrid-search",
     "ontology",