Fix parsing when startxref points near xref keyword

[vitormattos]

Bug

A valid PDF from the veraPDF corpus failed with:

Exception: Invalid object reference for $obj.

pdfinfo reports Pages: 1, so the document should be parsed successfully.

Root Cause

In this file, startxref points to a byte adjacent to the xref keyword (off by one).
The parser required an exact xref match at startxref, then incorrectly fell back to cross-reference stream parsing.
That path tried to resolve an invalid object reference and raised the exception.

Fix

• Normalize the startxref probe offset by skipping leading PDF whitespace.
• Tolerate the case where startxref points one byte inside xref (ref) by probing one byte back.
• Use the normalized offset consistently for deciding between xref table and xref stream parsing.

Tests

• Added regression fixture: samples/bugs/PullRequest794.pdf
• Added integration test: DocumentIssueFocusTest::testParseFileWhenStartxrefPointsNearXrefKeyword

The new test fails on master and passes with this patch.

Source PDF

• veraPDF corpus test file:
  https://raw.githubusercontent.com/veraPDF/veraPDF-corpus/master/PDF_A-2b/6.6%20Metadata/6.6.2%20Metadata%20streams/6.6.2.3%20Schemas/6.6.2.3.2%20Extension%20schemas/veraPDF%20test%20suite%206-6-2-3-2-t01-pass-c.pdf

[vitormattos]

Closing this as duplicate of #798 to keep discussion/history in a single PR.