Track parser and xref recovery parity with upstream PDFBox

[erikbra]

Parent: #650

Upstream issues

• https://issues.apache.org/jira/browse/PDFBOX-5032 - Refactor COSParser, especially xref parsing.
• https://issues.apache.org/jira/browse/PDFBOX-6201 - Xref brute-force recovery should keep valid entries.

Current PdfBox.Net assessment

PdfBox.Net has some adapted xref parsing and bad-offset recovery, but it does not yet have a clearly separate, mature brute-force xref repair layer equivalent to the direction requested upstream.

The port should avoid inventing an independent parser architecture before Apache PDFBox settles the upstream design. If a concrete upstream commit lands, port that shape as mechanically as possible.

Next actions

• Watch the upstream JIRA issues and corresponding Apache commits.
• If public fixtures are available, add them to the runtime parity corpus before changing behavior.
• Compare Java trunk and PdfBox.Net output for broken xref / incremental-save cases.
• Port/adapt confirmed upstream fixes rather than creating a divergent parser strategy.

Acceptance criteria

• Repro fixtures exist for the known xref repair failures, or the issue documents why no public fixture is available.
• PdfBox.Net behavior is compared against the matching Apache PDFBox branch.
• Any implementation follows upstream Apache PDFBox design once available.