Fix duplicate Markdown images across pages

[DresdenGman]

What this changes

• scopes Markdown image filenames by page for multi-page documents
• keeps the existing filenames for single-page results
• updates both Markdown references and the returned image mapping
• preserves page metadata when PaddleOCR-VL concatenates page results
• adds regressions for regular image blocks and images embedded in tables

Root cause

Image paths were derived only from the block label and coordinates. Two images at the same coordinates on different PDF pages therefore shared one dictionary key and output filename, causing the later image to overwrite the earlier one.

Validation

• two isolated regression tests covering same-coordinate images on two pages
• python3 -m compileall on all changed Python files
• Ruff checks and formatting checks on the new converter, mixin, and test changes
• git diff --check

The full PaddleX test suite was not run locally because this checkout is intentionally sparse and does not contain the complete optional runtime dependencies.

Fixes PaddlePaddle/PaddleOCR#18305

[CLAassistant]

All committers have signed the CLA.