Harden Python evaluation responses for pytest collection/import errors

[NickK21]

Summary

Python test runs can currently be reported incorrectly when pytest fails before normal test assertions run.

A common failure mode is:

• state: "passed"
• tests_run: 0
• non-empty runtime_error

This happens because the current Python result handling relies on regex parsing of pytest terminal output and does not model collection/import/syntax errors cleanly.

Current Problem

The Python execution path in codeval/executor.js does not distinguish well between:

• assertion failures
• import/collection errors
• syntax errors
• missing dependencies
• no tests collected

The response object is also missing structured fields that would make these cases easier to handle.

Planned Changes

• Add structured response fields for Python test execution:
  • errors
  • exit_code
  • no_tests_collected
• Ensure Python collection/import/syntax errors are reported as failed, not passed.
• Attach the subprocess exit code to Python execution errors in the executor.
• Add a synthetic failure_details entry for collection/import/syntax failures when no normal assertion failure is available.