A missing pytest-cov report reads as a green 0.0% Python coverage

[aclark-aquaveo]

Raised in review of #105 and deliberately left out of that PR, which closed one producer of a missing Python coverage report while leaving the detector blind.

A missing Python report is not just silent — it is green

• coverage_tools/coverage_generator.py:581 — py_raw = _py_percent_from_summary(py_summary) if py_summary.exists() else 0.0. A missing report is indistinguishable from a genuine 0%.
• generator_tools/ci_file_generator.py:140 — python_threshold defaults to 0. So 0.0 >= 0.0 → py_pass = True → exit 0, printing Python: 0.0% (threshold 0.0%) -> PASS for any library that has not set a threshold.
• coverage_generator.py:559-576 — three if X is not None: copies with no else, no warning, no log. Nothing distinguishes "copied the report" from "found nothing to copy". _find_pytest_cov_artifact's docstring justifies None as legitimate when pybind=False, but at this call site pybind is True by construction (:520-527), so None can only mean something broke.
• ci_templates/github-coverage.yaml.jinja:83,93 sets if-no-files-found: ignore (the default is warn), removing the last place an operator would see a red flag. gitlab-ci.yml.jinja:379-389 drops the Python link with no echo, and docs/USAGE.md:591 documents that as intended.

The C++ half does not accept this asymmetry: _assert_gcovr_collected_data (:392) raises a three-cause diagnostic on zero instrumented lines. The Python half has no equivalent — and _py_percent_from_summary (:459) raises on schema drift with a docstring saying it does so "rather than collapsing to an indistinguishable 0%", while the file being absent entirely collapses to exactly that.

#105 closed one route in (the Linux Debug pybind build skipped its wheel, so no Python tests ran). Every other route still lands green: XMS_COVERAGE not reaching the recipe, a cache hit with no build folder, a pytest-cov output path change.

Suggested scope

1. Raise on an absent cov-py-summary.json when the pybind coverage build was expected to produce one, with a diagnostic naming the candidate causes — mirroring _assert_gcovr_collected_data.
2. Log explicitly in the three if X is not None: arms when there was nothing to copy.
3. Reconsider the 0 threshold defaults, or separate "no threshold set" from "0% is acceptable" so a missing report cannot report PASS.
4. Revisit if-no-files-found: ignore in the GitHub coverage template.

Two smaller adjacent items

• xms_conan2_file.py:526-533 runs _save_test_artifacts() in a bare finally, so an unguarded rmtree/copytree failure there replaces a propagating test failure. Give the Debug pybind module the _d name its consumers link #105 makes run_python_tests() reachable on the coverage path, where CI sets XMS_TEST_ARTIFACTS_DIR.
• xms_conan2_file.py:350-356 is an except ConanException: pass on python_version propagation, so a typo'd override is discarded silently and later surfaces as No pybind=True... package found from _find_coverage_package, pointing at the wrong component.

No repo sets [ci].coverage = true today, so nothing is failing right now — the same "luck, not safety" that #105's second change was about.