Expose per-page pages_with_images signal

PdfClassification and PdfResult already expose per-page structural signals (pages_needing_ocr, layout.pages_with_tables, layout.pages_with_columns) so callers can make routing decisions without a second pass. One signal is conspicuously missing from that set: which pages contain embedded images. The internal detector already computes this (PageAnalysis.has_images) during its existing classification pass, but aggregates it into pdf_type and discards the per-page list before it reaches the public API.

This signal is useful for my very own pdf processing pipeline. Would love to make a contribution to add this.

Start at PageAnalysis.has_images and trace how the existing classification pass aggregates it into pdf_type before reaching PdfClassification and PdfResult. Confirm the per-page image list is preserved and exposed alongside the existing per-page signals, then validate the public result for PDFs with and without embedded images.

Written by the indexing model from the issue text.