Add /read-pdf skill with cached marker-based PDF-to-markdown conversion

[nsmiller2501]

Summary

This PR adds /read-pdf, a local-conversion alternative to /split-pdf for academic-paper ingestion.

What changed

• Adds a new /read-pdf skill that converts PDFs to markdown with marker-pdf.
• Adds helper scripts for one-time converter install and cached conversion.
• Caches conversions by SHA-256 under ~/.cache/claude-pdf-converter.
• Writes the same _text.md structured output contract used by /split-pdf: bibliographic metadata followed by research notes.
• Documents the marker-only backend, cache layout, GPU behavior, and failure mode.

Why

/split-pdf is useful for vision-based paper reading, but local markdown conversion can preserve equations, table structure, and figure references with much lower conversation-context cost after setup.

Testing

• Reviewed the branch diff against the PR notes.
• Re-audited the helper scripts and documentation after fixing portability/doc drift.
• Ran /usr/local/bin/python3 -m py_compile .claude/skills/read-pdf/convert.py .claude/skills/read-pdf/install.py.
• Did not run a live marker-pdf install/conversion smoke test in this PR branch.