feat(engine): wire Context variant types into detection operations

[martsokha]

Summary

Wire the existing context variant types (FaceData, VoiceData, EmbeddingData, PatternData, TextData, etc.) into detection operations so they serve as reference data during entity detection.

Design

Context entries are already loaded into the ContextCache via LoadContextOp and accessible on envelope.shared.context_cache. The missing piece is detection operations actually reading and using them:

• FaceData/VoiceData: biometric detection operations (CV, speaker ID) compare against known face/voice embeddings to identify specific individuals
• EmbeddingData: NER operation uses vector similarity search against reference embeddings for semantic matching
• PatternData: pattern recognition operation loads custom regex/dictionary patterns from context at runtime, extending the built-in pattern engine
• TextData/TagData: reference name lists and keywords used as deny/allow lists during detection

Tasks

• PatternRecognitionOp reads PatternData from context cache to extend pattern engine
• EntityRecognitionOp reads TextData/EmbeddingData from context for reference matching
• VisualExtractionOp reads FaceData from context for face matching
• AudialExtractionOp reads VoiceData from context for speaker identification
• Integration tests with context-loaded reference data

🤖 Generated with Claude Code