[codex] Isolate generated package proof tests

[rickardvh]

Summary

• Run heavyweight generated-package checker scenarios in short subprocess cases instead of the grouped pytest process.
• Preserve the existing static proof assertions while containing native failures to each subprocess case on Windows CPython 3.13.
• Archive the AW execution plan used to dogfood the issue.

Root cause

tests/test_generated_command_package_proof_runner.py repeatedly imported and exercised static proof helpers in-process. On Windows CPython 3.13 that grouped run could intermittently terminate with access violations even though the same proof commands passed as standalone processes.

Validation

• uv run pytest tests/test_generated_command_package_proof_runner.py -q
• make test-workspace
• make lint-workspace
• uv run python scripts/check/check_generated_command_packages.py

Dogfooding

Filed follow-up #1218 after closeout suggested an invalid agentic-workspace planning summary command.

Closes #1212