ci: test results and coverage as artifacts, and the integration suite under 5.1

[fadwen]

Summary

The tests job computed coverage and gated on it but wrote no file, and no job kept its test results: a failure had to be read out of the log, and coverage could not be compared between runs. The Windows PowerShell 5.1 job ran the unit suites only, so a 5.1 trap in the harness showed on a device before it showed in CI. Every test job now writes its results as NUnit XML and the tests job its coverage as JaCoCo XML, uploaded as artifacts whether the job passes or fails, and the 5.1 job runs the integration suite as well.

Stacked on #22 (base branch); the diff is this change only. Merge #16 through #22, then this. The workflow runs on pull requests against main, so this change is exercised when the stack lands.

Changes

• .github/workflows/quality-gates.yml. TestResult.OutputPath and CodeCoverage.OutputPath set before Invoke-Pester in the tests job, so a failed run still leaves both files; TestResult.OutputPath in the 5.1 and ARM64 jobs; an actions/upload-artifact@v7 step with if: always() after each. The 5.1 job's Run.Path gains ./Tests/Integration and writes the same step summary the ARM64 job does. The header comment says what each job covers.
• .gitignore. testResults-*.xml, next to the two names already there.
• Changelog. A Changed entry.

Verification

• The workflow parses as YAML (powershell-yaml 0.4.12): four jobs, every step with a name or a uses, three upload steps with if: always().
• actions/upload-artifact v7.0.1 is the current release, the same major as the actions/checkout@v7 the workflow uses.
• The workflow was dispatched on this branch, which carries the whole stack (fix(repair): Repair-IntuneScript takes -Context, -Architecture and -EnforceSignatureCheck #16 to ci: test results and coverage as artifacts, and the integration suite under 5.1 #23): run 37439917691, four jobs green. Windows PowerShell 5.1 with the integration suite: 964 passed, 47 skipped. Coverage 95.04%. Three artifacts uploaded: tests-and-analysis (results and coverage), windows-powershell-5-1, windows-on-arm.
• The first dispatched run, before the stack's last fixes, failed two jobs and both failures were real: a 116-character line in a test file of feat(harness): script paths from the pipeline, and a view for every result type #22, and the ARM64 runner's Hyper-V and installed modules against the inbox-module test of test(rules): the in-box module table is held against the Windows client the tests run on #21. Both are fixed in their own PRs; the artifacts from that failed run were uploaded as intended.

Notes

• If the integration suite proves slow or flaky under 5.1 on the hosted runner, the fix is in the suite or the harness, not in dropping the job back to unit tests: a test that passes under pwsh and fails under 5.1 is the finding this job exists for.

[fadwen]

Notes on the lines whose reason the diff does not show.

[fadwen]

Set on the configuration rather than written afterwards from $result, so the files exist the moment Invoke-Pester returns, before the two throws below can end the step; the upload runs with if: always() and finds them either way.

[fadwen]

Validation/Tests/Integration stays out: it needs a lab account and a Graph session the runner has not, and those files skip themselves; listing them would only add skipped rows.

[fadwen]

Three artifacts rather than one because the three jobs run in parallel on different runners and an artifact name may be uploaded once per workflow run.