Incrementally resume long-PDF ingestion using cached PageIndex doc_id

[plasma16]

Summary

• add long-PDF ingest checkpoint state in .openkb/long_pdf_jobs.json
• cache doc_id and description after successful PageIndex indexing
• on re-run, reuse cached doc_id for long PDFs and retry compilation directly
• persist index/compile failure state for troubleshooting and incremental retry

Why

When long PDF ingestion fails after indexing, re-running currently re-indexes the same document. This change makes retries incremental for long PDFs while leaving existing skip behavior unchanged for other file types.

Scope

• only long-document (long_pdf) ingestion path
• no queue/cursor behavior for non-PDF files

[KylinMountain]

Thanks @plasma16. In the default local mode, PageIndex already content-hashes each file and reuses the cached doc_id without re-running the expensive parse/index pass, so local re-runs do not re-index from scratch; the only remaining gap is cloud-mode retries. The patch also no longer applies after #142 rewrote the function it targets. Closing for now — appreciate the dig into the resume path.