Stream OCR result by page & code restructure

[kailingding]

Major Changes:

1. Code reorganization

   • Split large utils.ts file into smaller, focused modules under /utils directory
   • Moved OpenAI-related code to /models directory

2. OCR latency reduction by 4s-10s

   • previous process: preprocess all images -> run LLM calls in parallel
   • improved process: [preprocess one image -> run LLM call] in parallel

3. Error handling improvements

   • Added new ErrorMode enum (THROW or IGNORE)
   • Implemented retry mechanism with maxRetries parameter
   • Added error status tracking in page processing

4. Progress tracking

   • Added Page status tracking (SUCCESS/ERROR)
   • New summary object with success/failure counts
   • Added pre/post processing callbacks for better progress monitoring

Documentation Updates

• Added new parameters documentation in README
• Fixed markdown formatting issues
• Updated example outputs to reflect new response format

The changes primarily focus on code organization and latency enhancement during document processing.