fix(ci): raise test job timeout for instrumented main lane

[mldangelo]

Summary

• raise the test job timeout from 35 to 75 minutes so the main-branch Python 3.12 coverage lane can complete

Root cause

The main-only "fast tests with coverage" step measures the suite under branch coverage, which costs ~3x the uninstrumented lanes (measured locally: 16:36 instrumented vs ~6 min uninstrumented; the uninstrumented CI siblings take 22-26 min). The overhead concentrates in CPU-bound detector replay tests (e.g. tests/detectors/test_jit_script_detector.py, 1,818 tests) that compile fresh code objects per iteration, which defeats both the C tracer and the sysmon core (benchmarked 3.0x vs 2.7x — switching cores does not recover the budget).

Since the detector suites grew, every push to main has had Test Python 3.12 killed at the 35-minute cap (most recently run 27274096829, cancelled at 35m13s while the sibling 3.10/3.11/3.13 lanes passed in 22-26 min). The failure was previously masked because earlier jobs in the workflow failed first.

Why not alternatives

• COVERAGE_CORE=sysmon: benchmarked — only ~10% faster because the replay tests constantly execute fresh code objects; branch coverage also forces a fallback to the C tracer on Python ≤3.13.
• Excluding the detector file from instrumentation: loses its coverage data and still leaves a ~50 min lane; not worth the fidelity loss.
• The durable fix is reducing the intrinsic cost of the replay tests — left as a follow-up.

Validation

• Local: full fast suite under branch coverage completes with the same failure set as uninstrumented (no coverage-induced failures), 996s locally → projected ~55-65 min on a 4-vCPU runner, within the new 75-minute cap.
• Uninstrumented matrix versions are unaffected (they finish in ~25 min regardless of the cap).

🤖 Generated with Claude Code