test_crash_cells.py runs one module-scoped journey (install, the shared
gateway phase, six crash cells) inside a single file. It took 1,750-2,090 s
on normal runners (87% of the 2,400 s per-file budget) and was killed at
2,400 s on a runner that was uniformly 1.3-1.8x slower (run 37243653456:
32 passed, 0 failed, crash cells SIGKILL'd mid-journey). Raise the per-file
budget to 3,600 s and the job to 75 min so runner variance is not a verdict.
e9f7772602