[LauraGPT]

Summary

Add package markers to the MCP and OpenAI API example directories so their
standalone smoke_test.py scripts have distinct import names under pytest.
Both documented CLI paths remain unchanged. No model, inference, Docker
publication policy, pytest discovery setting or exclusion is changed.

Add a real subprocess regression for prepend, append and importlib collection,
plus both direct --help invocations. A small path-filtered, read-only CI job
runs this and the existing example metadata contracts on Python 3.10 and 3.12,
retaining exact source-commit, dependency and JUnit evidence.

Validation

• Before the fix: 2 failures (prepend/append import-file mismatch), 3 passing
  controls (importlib and both standalone help commands).
• After: 12 focused pytest cases pass, zero skips; the existing MCP unittest
  discovery command passes 21 cases. No live service/model/provider call.
• Bare offline python -m pytest -q was attempted, not silently narrowed. The
  example name collision is gone. Four pre-existing collection errors remain:
  runtime/python/utils/test_cer.py, test_rtf.py, test_rtf_gpu.py parse
  mandatory CLI arguments on import; the separate funasr_onnx dependency is
  missing for tests/test_funasr_onnx_paraformer_online_empty_final.py.
  Two existing optional skips remain. This is NOT a passing full-suite claim.
• All 4159 original blobs are unchanged. The repair consists of four new files.
  No GPU/runtime change, pretrained ASR quality result, Nano work, release or
  PyPI upload is implied. Actual code review is by the maintainer implementing
  this fix, not claimed as an independent GitHub review.
• Exact-head CI and downloaded evidence must pass before normal merge.
  No administrator bypass, self approval or protection change.