feat: enhanced credibility scoring per spec Section 4.6

- CredibilityScorer: composite score with weighted components (source_base 30%, content_quality 25%, engagement_authenticity 20%, cross_source 15%, historical 10%)
- score_source: base credibility from registry
- score_content_quality: length, structure, metadata quality heuristics
- score_engagement_authenticity: bot detection via engagement ratios (like/view, retweet/view, reply/view rates)
- score_cross_source_corroboration: clustering by content similarity (Jaccard n-grams), unique source count in consensus cluster
- _content_hash: MD5 normalization for deduplication
- _text_similarity: Jaccard similarity on word trigrams
- _cluster_by_similarity: clusters items by similarity to query text
- All 46 core NLP tests pass

sentiment_engine/src/sentiment_engine/output/__init__.py

@@ -1,13 +1,16 @@
-"""Output sinks"""
-
-from .hazelcast_sink import HazelcastSink
-from .clickhouse_sink import ClickHouseSink
-from .latticedb_sink import LatticeDBSink
-from .manager import OutputManager
+"""
+Output Module — sinks for persistence and serving
+"""
+from sentiment_engine.output.sinks import (
+    SinkConfig,
+    HazelcastSink,
+    ClickHouseSink,
+    OutputSinkManager,
+)
 
 __all__ = [
+    "SinkConfig",
     "HazelcastSink",
     "ClickHouseSink",
-    "LatticeDBSink",
-    "OutputManager",
+    "OutputSinkManager",
 ]