fix(workflow): add PDF extraction fallback

[luochen211]

Summary

Fixes #37488.

The workflow Document Extractor can currently return a successful but empty result for some valid PDFs because Dify delegates PDF text extraction to Graphon 0.5.x, whose PDF extractor only uses PDFium. This PR installs a small Dify-side fallback during workflow node registration: keep Graphon/PDFium as the primary extractor, but when it returns only whitespace for .pdf / application/pdf, retry the same bytes with pypdf.

This keeps existing non-empty extraction behavior unchanged, adds pypdf as a direct API dependency because Dify now imports it, and includes a unit test that patches the Graphon registry to reproduce the empty-PDFium case.

Refs langgenius/graphon#189.

Screenshots

N/A, backend workflow extraction behavior.

Checklist

• This change requires a documentation update, included: Dify Document
• I understand that this PR may be closed in case there was no previous discussion or issues. (This does not apply to typos!)
• I have added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
• I have updated the documentation accordingly. (N/A, no user-facing docs change.)
• I ran make lint && make type-check (backend) and cd web && pnpm exec vp staged (frontend) to appease the lint gods. Full command not run; scoped backend validation is listed below.

Validation

• uv lock --project api --check
• uv run --project api pytest -o addopts='' api/tests/unit_tests/core/workflow/test_document_extractor_pdf_fallback.py api/tests/unit_tests/core/workflow/test_node_factory.py (52 passed)
• uv run --project api --group dev ruff format --check api/core/workflow/document_extractor_pdf_fallback.py api/core/workflow/node_factory.py api/tests/unit_tests/core/workflow/test_document_extractor_pdf_fallback.py
• uv run --project api --group dev ruff check api/core/workflow/document_extractor_pdf_fallback.py api/core/workflow/node_factory.py api/tests/unit_tests/core/workflow/test_document_extractor_pdf_fallback.py
• uv --directory api run pyrefly check core/workflow/document_extractor_pdf_fallback.py core/workflow/node_factory.py (0 errors; existing warnings in node_factory.py)
• uv --directory api run mypy core/workflow/document_extractor_pdf_fallback.py core/workflow/node_factory.py --check-untyped-defs --disable-error-code=import-untyped

[github-actions]

Pyrefly Type Coverage

Metric	Base	PR	Delta
Type coverage	48.58%	48.58%	-0.00%
Strict coverage	48.09%	48.09%	-0.00%
Typed symbols	27,986	27,987	+1
Untyped symbols	29,920	29,923	+3
Modules	2892	2894	+2