ci(tests): the resource-monitor test can see a sample written mid-line (flaky) #1888

tests/project/test_resource_monitor_1740.py::TestTheSamplesReachTheTestsLog::test_samples_stream_into_the_tests_log_and_the_stop_step_ends_the_sampler failed on #1881's Python 3.12 job (117297298108), at test_resource_monitor_1740.py:177. The PR changes only docs/source/conf.py and docs tests, so the failure is unrelated to it.

The assertion checks that every sample line starts on its own line, even after pytest's unfinished progress line. One sample was written straight after the stub run's progress characters:

'uuu.u[monitor]    2438 692764 948892      02:05 /opt/hostedtoolcache/Python/3.12.15/x64/bin/pyt'

So the step that interleaves the sampler's output with the tests' output does not always start a sample on a new line. It depends on when the sample lands relative to pytest's writes, which makes it a timing race. The same test passes in local tests/project runs, including today's combined tree on main, so it is intermittent. A flaky blocking test turns any PR red at random.

Fix: make the start-of-line guarantee independent of timing in the workflow step and in the test's stub, or make the assertion tolerate the race only where the real step cannot cause it. Add a test that forces the interleaving deterministically, and check that it fails without the fix. The resource-monitor steps live in .github/workflows/experimental.yml.