feat(core): add PDF text extraction to read_file tool

[scrollDynasty]

Summary

Implements PDF text extraction support for the read_file tool.

Closes #1149

Problem

Previously, attempting to read a .pdf file returned binary content or an error about unsupported modality — making PDFs completely unusable in Qwen Code.

Solution

• Detect .pdf extension in processSingleFileContent
• Use pdf-parse to extract text content from the binary file
• Remove PDF from mediaModalityKey() — text extraction works regardless of model modality

Edge Cases Handled

• Normal PDF with text → returns extracted text
• Encrypted/password-protected PDF → clear error message
• Image-only/scanned PDF (no text layer) → suggests using OCR
• Corrupted or missing file → returns actual error

Testing

Manually verified before/after:

• Before: "I cannot directly read PDF files"
• After: successfully extracts and returns text content

Unit tests added:

• Successful PDF text extraction
• Encrypted PDF handling
• Image-only PDF handling
• Non-existent file handling

All checks pass:

• npm run test 113 tests pass
• npm run lint:ci
• npm run build
• npm run typecheck

[tanzhenxin]

Thanks for working on this, @scrollDynasty! PDF support is definitely a gap we need to fill.

We actually have an overlapping PR in #3160 that covers the same problem — making PDFs readable for text-only models. That one takes a slightly different approach (system pdftotext with fallback, preserving native PDF pass-through for multimodal models like Claude/Gemini, page range support, etc.), so we're going to move forward with that implementation.

Appreciate you taking the time to put this together — the error handling for encrypted and image-only PDFs is a nice touch. Going to close this one in favor of #3160.