fix: tolerate startxref offset inside xref keyword

[vitormattos]

Summary

• Fixes parsing when startxref points one byte inside the xref keyword (ref) or includes leading whitespace drift.
• Uses a normalized xref offset before deciding xref table vs xref stream path.
• Adds an integration regression fixture and test for the failing corpus file.

Reproduction

• Fixture parses with pdfinfo but failed in parser with Invalid object reference for $obj..
• New test: testParseFileWhenStartxrefPointsToLeadingWhitespaceInXrefStream.

Changes

• RawDataParser::getXrefData() now computes and uses a whitespace-adjusted startxref offset.
• Adds one-byte tolerance for malformed offsets that land on ref instead of xref.
• Adds fixture: samples/bugs/PullRequestStartxrefWhitespaceXrefStream.pdf.
• Adds regression assertion in integration tests.\n\nPDF Source
• Corpus: veraPDF-corpus
• Fixture origin: PDF_A-1b/6.1 File structure/6.1.2 File header/veraPDF test suite 6-1-2-t01-fail-a.pdf
• Source URL: https://github.com/veraPDF/veraPDF-corpus/blob/staging/PDF_A-1b/6.1%20File%20structure/6.1.2%20File%20header/veraPDF%20test%20suite%206-1-2-t01-fail-a.pdf
• Local copy used in tests: samples/bugs/PullRequestStartxrefWhitespaceXrefStream.pdf

[vitormattos]

Added an extra regression sample from the same investigation: PullRequest794.pdf (veraPDF 6-6-2-3-2-t01-pass-c). This complements existing coverage and confirms the same startxref tolerance fix across another real-world file.

[vitormattos]

This standalone PR has been restacked into the RawDataParser consolidation chain to reduce conflict hotspots in shared test files.\n\nSuperseded-by chain:\n- upstream base: #796\n- fork stack (PR797 replay): https://github.com/vitormattos/pdfparser/pull/26\n- fork stack (this PR798 replay): https://github.com/vitormattos/pdfparser/pull/27\n\nClosing this standalone PR to keep a single merge path per source-file group.