feat: Add document chunking for large PDFs (300+ pages, 200+ MB)

[majcheradam]

Problem

Users need to process large documents (300-500 pages, 200-300 MB) but current OCR services have size limits:

• Mistral OCR: max 50 MB, 1000 pages
• Most vision models struggle with very large documents

Proposed Solution

Add document chunking/splitting functionality as a pre-processing step:

Core Features

• Split large PDFs by page count (configurable, e.g., 100 pages per chunk)
• Split large PDFs by file size (e.g., max 50 MB per chunk)
• Async processing of chunks with result aggregation
• Progress tracking per chunk via WebSocket

API Design

const job = await client.jobs.create({
  file: largePdf,
  type: "parse",
  chunking: {
    enabled: true,
    maxPages: 100,      // or
    maxSizeMb: 50,
  }
});

Implementation Notes

• Use pdf-lib or similar for splitting without re-rendering
• Queue each chunk as sub-job
• Aggregate results maintaining page order
• Consider batch processing discount (Mistral: 50% cheaper in batch mode)

References

• Mistral Document Chunking Cookbook
• LlamaIndex Chunking Strategies

Use Case

Customer inquiry about processing 300-500 page documents (~200-300 MB)