And indeed, after removing the `grep ERROR` part of the repro, even the repro seems to be invalid:

```
cpython main $ ./python.exe -m test.test_importlib -v -k test_entry_points_unique
test_entry_points_unique_packages (test.test_importlib.test_metadata_api.APITests)
Entry points should only be exposed for the first package ... ERROR
test.test_importlib.test_windows (unittest.loader.ModuleSkipped) ... skipped "No module named 'winreg'"

======================================================================
ERROR: test_entry_points_unique_packages (test.test_importlib.test_metadata_api.APITests)
Entry points should only be exposed for the first package
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/jaraco/code/public/cpython/Lib/test/test_importlib/test_metadata_api.py", line 97, in test_entry_points_unique_packages
    raise ValueError("Failing on purpose")
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: Failing on purpose

----------------------------------------------------------------------
Ran 2 tests in 0.006s

FAILED (errors=1, skipped=1)
```

I'm effectively unable to replicate the behavior reported by Brett in the other issue (unless `| grep ERROR` is consider an important use-case to support).

At this point, I propose one of two options:

1. Back out PR 30194 and restore the prior behavior with descriptions.
2. Leave PR 30194 to suppress descriptions for the default runner.

And then simply declare that docstrings are acceptable for CPython tests.

Given that descriptions are on by default, I'm slightly inclined toward (1).

Thoughts?