research(editor): add a representative source-editing fixture corpus

[McanKul]

Goal

Build a small, license-safe fixture corpus for deciding which existing PDF text and image structures OffPDF can edit without corrupting the document.

This is a foundation for #11 and #12. It does not implement production editing.

Scope

• Add deterministic synthetic fixtures for common text structures:
  • Tj and kerned TJ arrays
  • subset CID fonts with and without usable ToUnicode maps
  • rotated or skewed text
  • Type 3 fonts
  • text inside nested Form XObjects
• Add deterministic synthetic fixtures for image structures:
  • a unique top-level Image XObject
  • an Image XObject reused on multiple pages
  • an image inside a Form XObject
  • an inline image
  • an image with a mask or soft mask
• Include CropBox/MediaBox offsets, /UserUnit, and 0/90/180/270 degree rotations.
• Document how every fixture was generated and what capability or failure mode it represents.
• Prefer generation scripts over committing large binary files.

Acceptance criteria

• Every normal fixture passes qpdf --check; intentionally malformed fixtures are clearly named and documented.
• Fixtures are synthetic, contain no sensitive or proprietary source documents, and have clear license provenance.
• The corpus opens in the existing pdf.js preview and backend test harness.
• A machine-readable manifest records the expected text/image structure and whether it is intended to be editable or unsupported.
• Tests verify fixture generation is deterministic and enforce a bounded total fixture size.
• Spike: evaluate safe editing of existing PDF text and images #11 can use the corpus to publish a compatibility matrix and implementation recommendation.

Out of scope

• Replacing source text or images.
• Claiming that a fixture is editable before Spike: evaluate safe editing of existing PDF text and images #11 establishes a defensible capability contract.