[CI] Fix chronic timeouts in Jenkins Code coverage stage

[bogdan-petkovic]

Motivation

The Jenkins Code coverage stage in mlir/utils/jenkins/Jenkinsfile has been chronically failing on PR builds and merges to develop, but the failure has been invisible for a long time:

• The whole stage runs inside a 60-minute activity-based timeout.
• Two parts of the pipeline produce little or no output that Jenkins sees:
  • lit buffers progress for the duration of ninja check-rocmlir (~60 minutes on the coverage matrix codepaths) and only flushes at the very end.
  • llvm-profdata merge -sparse operates on ~125 GB of *.profraw and runs silently for several minutes.
• The activity timer fires before llvm-profdata merge finishes → coverage_<cpath>.lcov is never produced → the Codecov uploader never runs → archiveArtifacts ... onlyIfSuccessful: true skips the report/lcov/html artifacts.
• The outer catchError(buildResult: 'SUCCESS', stageResult: 'FAILURE') plus an inner try { ... } catch (e) { echo "NOTE: Code coverage stage had an error or timeout: ..." } swallow the FlowInterruptedException → the build is green and the matrix bodies show ✓ even though no LCOV reached codecov.io.

Concrete evidence collected while investigating https://github.com/ROCm/rocMLIR/pull/ (the test-exclusion change):

[2026-05-20T12:09:03.701Z] Total Discovered Tests: 919
[2026-05-20T12:14:07.193Z] Sending interrupt signal to process
[2026-05-20T12:14:13.284Z] Terminated
[2026-05-20T12:14:13.333Z] script returned exit code 143
                            Timeout has been exceeded
[2026-05-20T12:14:13.572Z] NOTE: Code coverage stage had an error or timeout:
[2026-05-20T12:14:13.572Z] org.jenkinsci.plugins.workflow.steps.FlowInterruptedException: Timeout has been exceeded

The same FlowInterruptedException pattern was confirmed on a recent merge to develop (build from 2026-05-16), so codecov.io has been displaying a stale snapshot from whichever older build last produced a successful upload. The biweekly Teams report (.github/workflows/codecov-report.yml) reads totals.coverage from that stale data.

Technical Details

Single-line change inside the body stage of collectCoverageData(...) orchestration in mlir/utils/jenkins/Jenkinsfile:

-                                                                timeout(time: 60, activity: true, unit: 'MINUTES') {
+                                                                timeout(time: 180, unit: 'MINUTES') {

• Switch from activity-based to wall-clock timeout — removes spurious firings caused by lit's buffered output and llvm-profdata merge's silent runtime.
• Bump the budget to 180 minutes: ~60 min for ninja check-rocmlir + several minutes for llvm-profdata merge over ~125 GB + three llvm-cov invocations (report, export, show) + uploader download and upload, with comfortable margin.
• A short comment is added inline to explain why activity: was dropped, so this isn't reintroduced inadvertently in a future cleanup.

Test Plan

• Codecov on PR CI

Test Result

• PR CI with Codecov

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.