fix: recover PDF text after inline images

[he-yufeng]

Summary

• add a lazy PyMuPDF text-recovery pass for PDFs with inline image operators
• keep PyMuPDF optional: no default dependency change, and the fallback only runs when the package is installed
• warn instead of silently returning likely partial text when inline-image recovery is needed but PyMuPDF is unavailable
• add regression coverage for both the recovery and warning paths

Addresses #1870.

Test plan

• python -m pytest packages/markitdown/tests/test_pdf_memory.py -q -k "inline_image"
• python -m pytest packages/markitdown/tests/test_pdf_memory.py packages/markitdown/tests/test_pdf_tables.py -q
• python -m py_compile packages/markitdown/src/markitdown/converters/_pdf_converter.py packages/markitdown/tests/test_pdf_memory.py
• python -m mypy --ignore-missing-imports packages/markitdown/src/markitdown/converters/_pdf_converter.py packages/markitdown/tests/test_pdf_memory.py
• python -m pip install -e "packages/markitdown[pymupdf]"
• git diff --check

[he-yufeng]

Rebased onto current main; no conflicts.

Focused validation after the rebase:

python -m py_compile packages\markitdown\src\markitdown\converters\_pdf_converter.py packages\markitdown\tests\test_pdf_memory.py
$env:PYTHONPATH='C:\dev\GITHUB-clean\markitdown-1889\packages\markitdown\src'; python -m pytest packages\markitdown\tests\test_pdf_memory.py -q
git diff --check origin/main..HEAD

Result: 8 passed, 2 skipped for the focused PDF memory tests.