New OCR & Embedding Providers

Added

• DatalabOCR: New OCR processor using Datalab API with marker model
  • Supports three processing modes: fast, balanced, and accurate
  • Page range filtering and max pages limiting
  • Image extraction with optional captions
  • Cost tracking per page processed
• OpenAIEmbedder: New embedder using OpenAI's embedding models
  • Support for text-embedding-3-small (1536 dimensions)
  • Support for text-embedding-3-large (3072 dimensions)
  • Normalized vector embeddings with L2 norm ≈ 1
  • Token usage tracking for embedding operations
• Comprehensive integration tests for both new components
  • Regular functionality tests
  • Behavior tests ensuring embedding quality and OCR accuracy
  • Validation of embedding dimensions, normalization, and similarity properties
• Updated README with examples for DatalabOCR and OpenAIEmbedder
• Added section on using alternative OCR and embedding providers