Add integration test for full pipeline with LLM cache fixtures

[nicpottier]

Summary

Refactored the CLI to extract core pipeline orchestration into a reusable runPipeline() function, eliminating logic duplication between the CLI and tests. The CLI is now a thin wrapper that translates progress events to terminal UI.

Created an integration test that runs the full pipeline against raven.pdf (pages 1-3) with pre-populated LLM cache fixtures for reproducible, fast execution (~5 seconds with no API calls).

Details

• pipeline.ts: New file with runPipeline() and RunPipelineOptions interface
• cli.ts: Refactored to be a thin CLI wrapper; all pipeline logic moved to pipeline.ts
• index.ts: Exports runPipeline and RunPipelineOptions
• pipeline-integration.test.ts: Integration test validating all 6 pipeline steps end-to-end
• fixtures/raven-cache/: Git-tracked LLM cache fixtures (7 JSON files, ~3.5KB total)

Cache regeneration is simple: delete fixtures/raven-cache/ and rerun test with OPENAI_API_KEY set.

Tests

All 272 tests pass. The integration test confirms all pipeline steps complete successfully:

• PDF extraction (3 pages)
• Metadata extraction (title + metadata)
• Text classification (groups per page)
• Image classification (filtered by size)
• Page sectioning (sections per page)
• Web rendering (HTML per section)