test(pdf): regenerate llncsdoc snapshot (duplicate code blocks removed)

The nested-code dedup + language-label consumption also fixes llncsdoc, which had
the same doubly-detected code blocks: `\begin{document}`, `\institute{…}`,
`\url{…}` and `\begin{abstract}` each appeared twice and now appear once (verified
no content loss — every block is still present, just not duplicated). Regenerated
against the ONNX models; PDF snapshot conformance is back to fully exact.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01H3AUFjF71HP2YY9cD32wx8

Author: claude