Keep blank PDF pages as chunks to preserve page/chunk alignment

[Adityav369]

Problem

Since #399, pages that render as visually blank are dropped during ColPali PDF/office rendering. That compacts the chunk list, so chunk numbers no longer match source page numbers for any document containing blank pages — anything that maps a chunk back to its page (viewers, citations) points at the wrong page.

Change

• Render every page: blank pages become normal chunks again, restoring the chunk-N == page-N invariant.
• A page that fails to render now yields a white placeholder chunk instead of being silently dropped (a dropped page shifts numbering the same way a blank one did).
• Documents where every page fails to render still raise, and the pdf2image / text-extraction fallbacks are unchanged.
• Removed the now-unused blank-detection helpers.

Tests

• Updated rendering tests: all pages kept, placeholders substituted at the right positions, all-pages-failed still raises.
• Verified end-to-end that a PDF with interleaved blank pages produces one chunk per page, in page order, through both the normal and high-density batched paths.

🤖 Generated with Claude Code