fix(render): retain source tile indices for PDF subsets

[OrdoAbChao7]

Description

Preserve selected PDF pages' zero-based source tile indices. A full render maps page 3 to tile_0002.jpg, but pages=[3] previously wrote that page as tile_0000.jpg: pdf2image returns a range starting at page 3, while enumerate(images) restarted at zero.

Fix

Start enumeration at first_page - 1 and use idx + 1 for membership in the one-based page selection. Full-document tile identifiers and source-page mapping remain unchanged. Previously rendered subsets starting after page 1 need regeneration with their associated index metadata to adopt the corrected indices.

Regression tests

Eight offline cases cover full rendering, first/later/final pages, contiguous and sparse subsets, and unsorted input. They verify tiles.json, chunks.json, actual filenames, and selected image bytes against corresponding full-render tiles. Original Poppler integration tests are unchanged.

On upstream c4a0f53052306e009a024a2e344ede8aa5eda407: 5 failed, 3 passed (page 3: expected tile_0002.jpg, got tile_0000.jpg). With the fix: 8 passed. Removing the enumeration offset reproduces five failures; restoring it passes. A separate before/after full-render comparison found identical filenames, manifests, chunk metadata and mock-generated JPEG bytes.

Validation

Windows / Python 3.13.7, with PYTHONPATH=render/src:

• python -m pytest tests/test_pdf_page_indices.py tests/test_pdf_manifest.py tests/test_chrome_paths.py tests/test_network_idle.py -q -p no:cacheprovider — 22 passed, 1 skipped. Original PDF integration module skips because pdf2image is unavailable.
• uvx ruff check . — passed.
• uvx ruff format --check . — passed.
• git diff --check — passed.

Context

Focused PDF subset fix discussed in the maintainer review of #140, preserving the existing zero-based full-document layout. No linked issue.

[vercel]

@OrdoAbChao7 is attempting to deploy a commit to the andylizf's projects Team on Vercel.

A member of the Team first needs to authorize it.