Add automatic embedding generation with hybrid search capability

• Add automatic embedding generation when PDF has >5 chunks
  • Create unique Qdrant collections per document (user_{id}doc{id})
  • Store embeddings with rich metadata (document_id, chunk_id, position, etc.)
  • Add question parameter to upload API for user context
  • Implement local PDF processing mode when S3 unavailable
  • Add detailed logging for embedding storage and search similarity
  • Fix Qdrant vector ID format (use numeric IDs instead of strings)
  • Enable hybrid search on document-specific collections
  • Return all chunks to Java service for RAG processing

Closes IA-436