fix(test): repair orchestration checkpoint-resumer rot + workflow_memory state bleed (#11754)

[mrveiss]

Closes #11754

Thinking Path

Five pre-existing orchestration failures (baselined during #6828). Two root causes: the #6827 refactor renamed the checkpoint delegation seam (_checkpoint_manager → _checkpoint_resumer) and three spies still pointed at the deleted attribute; and the process-global circuit-breaker singleton accumulated failures from execution_modes_test.py, opening the workflow_step_execution breaker so two workflow_memory tests hit CircuitBreakerOpenError in group runs (confirmed with a minimal 2-file repro).

What Changed

• orchestration/workflow_executor_checkpoint_test.py: 3 spies re-pointed to executor._checkpoint_resumer.save — the behavioral assertions were already correct against the delegation seam.
• New orchestration/conftest.py: autouse fixture calling get_circuit_breaker_manager().reset_all_circuit_breakers() (the established reset seam) before every orchestration test — no reordering, no flaky marks; every test starts with breakers CLOSED.

Verification

• Before: 3 failed standalone (AttributeError) + 2 failed in group runs; full group 5 failed / 708 passed.
• After: full group (tests/orchestration/ + orchestration/ + services/mesh_brain/) 713 passed, 0 failed; polluter pair 60 passed (agent). Independent re-run: the three repaired files 63 passed.
• flake8 clean.

Model Used

Claude Fable 5 (subagent: general-purpose)

[github-actions]

✅ SSOT Configuration Compliance: Passing

🎉 No hardcoded values detected that have SSOT config equivalents!