fix: detect headings in PDF conversion via font-size analysis

[octo-patch]

Fixes #1640

Problem

When converting PDFs (especially tagged or structured PDFs with multiple heading levels), MarkItDown produced flat plain text with no heading markers. Users lost the document structure — all text was rendered at the same level regardless of whether it was a heading or body text.

Solution

Added _extract_text_with_headings() which uses pdfminer's extract_pages layout API to inspect per-character font sizes. The function:

1. Collects every text block and its dominant font size.
2. Determines the body text size as the statistical mode of all font sizes.
3. Any block whose font size is ≥ 15 % larger than the body size is treated as a heading.
4. Up to six distinct heading sizes are mapped to # … ###### (largest → H1).
5. Falls back to pdfminer.high_level.extract_text() on any error.

Plain-text PDFs (no form/table pages detected by pdfplumber) now use _extract_text_with_headings() instead of the bare pdfminer.high_level.extract_text() call, so heading structure is preserved where font size information is available.

Testing

• All existing PDF tests pass: test_pdf_masterformat.py, test_pdf_tables.py, test_pdf_memory.py.
• New test_pdf_headings.py covers:
  • PDF with 24 pt H1 and 18 pt H2 produces # … and ## … markers.
  • Body text (12 pt, the mode) is not converted to a heading.
  • Uniform-font PDFs (all same size) produce no heading markers.
  • Function always returns a non-empty string.