fix: capture Python tool execution output

[Jarvis636431]

Summary

Fixes python_tools.run_python_code so agents receive output produced by executed Python code instead of only a generic success message.

Closes #105.

Changes

• Capture and return standard output.
• Capture standard error and mark it with a [stderr] prefix.
• Preserve variable_to_return and no-output behavior.
• Preserve explicitly supplied empty execution namespaces.
• Add focused unit tests for stdout, stderr, combined output, and namespaces.

Validation

• python -m pytest tests/unittest/test_tool/test_python_tools.py -q — 7 passed.
• ruff format --check oxygent/preset_tools/python_tools.py tests/unittest/test_tool/test_python_tools.py — passed.
• docformatter --check oxygent/preset_tools/python_tools.py — passed.
• Full unit suite: 342 passed, 13 skipped, 5 unrelated failures because train-ticket tests require access to 12306.cn, which was unavailable in the test environment.