fix #426: Flash reads AES PDFs and guards lazy PyPDF2 failures

[xinyao-li]

Add PyCryptodome as PyPDF2's AES backend and force page-tree resolution inside the existing guarded open so an unavailable secondary parser falls back to PDFium instead of crashing.

Summary

• add PyCryptodome as PyPDF2's AES backend
• eagerly resolve the PyPDF2 page tree inside the existing guarded open
• fall back to the PDFium-only channel if PyPDF2 cannot read the document

Fixes #426.

Root cause

PyPDF2 parses the page tree lazily. PdfReader() succeeded inside the
existing try/except, but len(reader.pages) later raised DependencyError
outside that guard while decrypting an AES-encrypted object stream.

Resolving and caching the page count in _PdfDoc.__init__ keeps that lazy
failure inside the existing guarded open used by both sequential and
parallel parsing.

Test plan

• reproduced the original failure with an AES-128, empty-user-password PDF
• verified AES-128 and AES-256 PDFs parse after installing PyCryptodome
• verified PDFium-only fallback without PyCryptodome in sequential and
  parallel parsing
• added regression tests for the lazy PyPDF2 failure
• 386 passed, 53 skipped