Improve visibility into PDF extraction results

[Parsenal]

Problem

PDF ingest currently produces a final extracted text payload, but the UI does not make it easy to inspect what happened during the Extract step. Users can view the original PDF, and source detail exposes extracted passages, but there is no focused place to compare the PDF with the extracted text or see which extraction path was used.

This makes it harder to debug extraction quality, compare Docling vs fallback behavior, and build confidence before compilation.

Proposed first step

Add a lightweight extraction preview for PDF sources:

• make the Extract step on the source detail page open an extraction preview
• show the original PDF beside the stored extracted text
• expose small extraction metadata, including the engine used and page count when known
• keep existing PDFs compatible by showing unknown metadata when the source was ingested before the fields existed

Non-goals for the first PR

• no page-level extracted text storage yet
• no re-extraction/backfill workflow yet
• no Docling-specific full conversion artifact storage yet

Those can be follow-up issues once the lightweight preview establishes the user-facing workflow.

[Parsenal]

Draft sample PR opened: #830. It implements the lightweight extraction-preview path and keeps the PR in draft until the generated OpenAPI file can be added to the branch.