feat(jeparag): add native @magek/jeparag package for hybrid RAG and ambient agents

[emezac]

Summary

This PR introduces the @magek/jeparag package into the Magek monorepo under packages/jeparag. This PR is related to #808 issue.

@magek/jeparag brings Joint Embedding Predictive Architecture (JEPA) dense vector representations combined with Okapi BM25 sparse keyword retrieval using Reciprocal Rank Fusion (RRF) (k=60).

Key Features

• DocumentLoader: Sliding window semantic text chunking (256 words, 50 overlap).
• BM25Index: Pure Okapi BM25 sparse keyword ranking engine.
• JEPAEmbedder: Multi-provider embedding engine (Google Gemini, OpenAI, L2 normalized hash fallback).
• ReciprocalRankFusion (RRF): Fusion algorithm combining dense vector and sparse keyword rankings.
• Event-Sourced Events & ReadModel: DocumentIngestedEvent, ChunkEmbeddedEvent, JeparagQueryProcessedEvent, and JeparagHybridSearchReadModel.
• JeparagAmbientAgent: Proactive background agent for automated document ingestion, indexing, and contextual hybrid RAG responses.
• Executable Demo: Included examples/hipaa-audit-demo.ts (rushx demo) demonstrating zero-recall-loss HIPAA compliance auditing.

Verification

• Unit Tests: npm test in packages/jeparag passed (5/5 passing).
• Monorepo Build: rush build completed with 100% success across all 16 workspace projects.

[emezac]

Reference: https://medium.com/@emezac/i-built-a-rag-system-with-jepa-encoders-heres-what-i-learned-3b5e7effc043