Flaky: test_initialize_output_dict fails intermittently under xdist (order-dependent)

[alongd]

arc/scheduler_test.py::TestScheduler::test_initialize_output_dict fails intermittently in CI, on main and on PRs that do not touch the scheduler at all.

FAILED arc/scheduler_test.py::TestScheduler::test_initialize_output_dict - AssertionError: False is not true

The assertion is self.assertTrue(self.sched1._does_output_dict_contain_info()) (arc/scheduler_test.py:350).

Evidence that it is order-dependent, not a real defect

• On main: the nightly run of 2026-08-10 (4a63d1aa) failed on this test — https://github.com/ReactionMechanismGenerator/ARC/actions/runs/31349161224 — with green nightlies on 08-11, 08-12 and 08-13 either side of it, and no relevant change in between.
• On an unrelated PR: Stop YAML from coercing a bare SMILES like NO into a boolean #981 (yaml_bool_smiles_fix) failed the same way — https://github.com/ReactionMechanismGenerator/ARC/actions/runs/31736868393/job/94570512325 — reported by gw4 at 1% of the run. That branch only touches arc/common.py and arc/species/species.py (plus their tests) and never goes near the scheduler.
• Locally: running the test on its own reproduces the failure on a checkout of main's content:

  HOME=$(mktemp -d) python -m pytest arc/scheduler_test.py -q -n0 -k test_initialize_output_dict
  → 1 failed, 42 deselected
  

  so the test passes only when earlier tests in the class have run first. Under pytest-xdist the shard that receives it may not have run them, which is why it surfaces sporadically rather than always.

Suspected cause

self.sched1 is built in setUpClass, and _does_output_dict_contain_info() reads state that other tests in the class populate. A likely contributor is global mutation of default_job_types: initialize_job_types() aliases and dels the module-level global, so a test that calls it changes what a later Scheduler sees. PR #926 carries a commit "Stop initialize_job_types from mutating the global default_job_types", which may remove the cause; worth re-checking this test's stability once that lands.

Suggested fix

Make the test self-sufficient rather than dependent on sibling tests — populate the output dict it asserts on inside the test (or in setUp), so it passes under -p no:randomly, in isolation, and on any xdist shard.