fix: recover hybrid xref offsets in pdf.js issue17147

[vitormattos]

Summary

• add regression fixture samples/bugs/PullRequest804-pdf.js.pdf from pdf.js issue17147
• add integration test covering parsing when a hybrid trailer points to /XRefStm and /Prev
• harden xref parsing to recover when startxref points near, but not exactly at, the xref keyword
• avoid type errors when xref-stream lookup starts on trailer dictionaries and follow trailer offsets instead

Root cause

issue17147.pdf contains a hybrid-reference structure where trailer metadata references both /XRefStm and /Prev, and one xref offset points near xref table entries rather than exactly at xref.

The parser raised a type error (array passed as object reference) and then failed with Unable to find xref.

Fix details

• in decodeXref(), follow /XRefStm trailers so hybrid references are decoded
• in decodeXrefStream(), guard object-reference type and recover from trailer dictionaries by following /XRefStm and /Prev
• in getXrefData(), search a small backward window for xref when offsets land shortly after the keyword

Validation

• red/green regression on tests/PHPUnit/Integration/DocumentIssueFocusTest.php with filter testParseFileWithArrayXrefObjectReferenceInStream
• full run of tests/PHPUnit/Integration/DocumentIssueFocusTest.php
• direct parse smoke-check of issue17147.pdf returns 1 page

All validation was run in Docker (php:8.3-cli) because host PHP is unavailable.

Source PDF

• https://github.com/mozilla/pdf.js/blob/master/test/pdfs/issue17147.pdf

[vitormattos]

Superseded by the RawDataParser consolidation chain in the fork.

This fix (hybrid xref offset recovery, XRefStm handling, backward xref scan) is included in vitormattos#32, stacked on fix/invalid-object-reference-tolerant-parser.