feat: add pytorch loader for pickle opcode security analysis

[GangGreenTemperTatum]

Summary

• Adds a new pytorch loader that statically scans PyTorch .pt/.pth checkpoint files for malicious pickle opcodes without executing anything
• Detects dangerous opcodes (GLOBAL, REDUCE, BUILD, etc.), classifies global imports as known-safe/unknown/suspicious, validates ZIP archive structure, and flags path traversal in zip entries
• Adds view_extra() dispatcher to view.py with pytorch-specific rendering (file structure, pickle analysis, global imports, color-coded findings)

Test plan

• poetry run ruff check dyana — lint clean
• poetry run ruff format --check dyana — format clean
• poetry run mypy --ignore-missing-imports --no-error-summary dyana — type check clean
• poetry run pytest dyana -v — 187 tests pass (20 new pytorch tests + 6 new view_extra tests)
• CI pipeline passes

🤖 Generated with Claude Code

Generated Summary:

• Updated the README to reflect that PyTorch checkpoints are now supported for profiling in Dyana.
• Added new functionality for analyzing PyTorch checkpoint files, which includes:
  • A new loader for PyTorch files that checks file structure and scans for suspicious pickle opcodes.
  • Ability to handle both ZIP format and legacy pickle files.
  • Various safety checks for dangerous opcodes within the pickle data.
• Incorporated a new command-line interface for analyzing PyTorch files, including usage examples.
• Added comprehensive tests around PyTorch loading and analysis, ensuring error handling and detection of known safe and dangerous imports.
• Minor code cleanups and refactoring for improved readability and maintainability, particularly in error handling messages and logging outputs.

This summary was generated with ❤️ by rigging