feat(pdf): extract bookmarks as markdown headings for hierarchical parsing

[r266-tech]

Summary

Implements Phase 1 of #393 — PDF bookmark/outline extraction and injection as markdown headings.

Problem

When parsing large structured PDFs (textbooks, pharmacopoeias, standards documents), the local pdfplumber strategy only outputs <!-- Page N --> HTML comment markers. Since MarkdownParser._find_headings() ignores HTML comments, it finds zero headings and falls back to paragraph-based splitting. This produces hundreds of flat numbered files (name_1.md through name_538.md) with no semantic organization.

Solution

• Added _extract_bookmarks() method that extracts PDF outlines via pdfminer's doc.get_outlines() (accessed through pdfplumber)
• Bookmark destinations are mapped to page numbers by resolving page object IDs
• Bookmarks are injected as markdown headings (#, ##, etc.) at the correct page positions before each page's text content
• MarkdownParser's existing heading-based splitting then naturally builds a hierarchical directory structure

Before

data/viking/resources/pharmacopoeia/
├── pharmacopoeia_1.md      # Page 2-3 raw text
├── pharmacopoeia_2.md      # Page 4 raw text  
├── ...
└── pharmacopoeia_538.md    # Last chunk

After (for PDFs with bookmarks)

data/viking/resources/pharmacopoeia/
├── 第一部_药材/
│   ├── 川木通.md
│   └── ...
├── 第二部_化学药/
│   └── ...
└── ...

Details

• Heading levels capped at 1–6 for markdown compatibility
• Empty/whitespace titles are skipped
• Unresolvable page destinations gracefully produce page_num=None (not injected at wrong position)
• New bookmarks_extracted metadata field added
• Fully backward-compatible: PDFs without bookmarks behave identically to before

Tests

7 unit tests added in tests/parse/test_pdf_bookmark_extraction.py:

1. Normal extraction with page mapping
2. No outlines → empty list
3. Missing get_outlines method → empty list
4. Empty/whitespace titles skipped
5. Level capping at 6
6. Unresolved page destinations
7. Exception handling (corrupt PDF)

Future Work (from #393)

• Phase 2: Font-size heading detection (fallback for PDFs without bookmarks)
• Phase 3: Directory auto-grouping (MAX_CHILDREN_PER_DIR threshold)

Ref: #393