fix(tools): drop exec-level OOMPolicy from systemd scope argv

[UltraInstinct0x]

What does this PR do?

Scope units only accept cgroup resource properties. _systemd_scope_argv added --property OOMPolicy=kill, which systemd 249 rejects on scopes with Unknown assignment (rc=1). That failed the availability probe in _systemd_run_user_scope_available, so restart_safe_gateway_child_argv failed closed and blocked every gateway child: kanban workers (spawn_failed x N, tasks going blocked) and cron workers (hundreds of failure streaks, silent automations). Removing the exec-level property restores probe rc=0 while keeping MemoryAccounting and MemoryMax isolation.

Related Issue

No issue filed. Happy to link one if maintainers prefer an issue first.

Type of Change

• Bug fix (non-breaking change that fixes an issue)

Changes Made

• tools/process_registry.py: drop OOMPolicy from _systemd_scope_argv with a note explaining why scopes reject it
• tests/tools/test_process_registry.py: update spawn contract (OOMPolicy must be absent) and add scope-validity invariant test

How to Test

1. On systemd 249 (Ubuntu 22.04): systemd-run --user --scope --quiet --unit=probe --collect --property=MemoryAccounting=yes --property=MemoryMax=1G -- /bin/true returns rc=0. Adding --property OOMPolicy=kill returns rc=1 with Unknown assignment.
2. scripts/run_tests.sh tests/tools/test_process_registry.py tests/cron/test_restart_safe_worker.py: 122 passed, 0 failed.
3. Live: after the patch plus gateway restart, a kanban worker assigned to a profile spawned successfully (previously spawn_failed with scope unavailable).

[alt-glitch]

This was generated by AI during triage.

Duplicate of #102357 — same fix (drop OOMPolicy=kill from the systemd --scope argv in tools/process_registry.py) for the same probe-fails-closed bug tracked in #102486. Other open competitors: #102655, #103762, #103768 (version-gated variant), #103930. Leaving open for a maintainer to pick a canonical PR.

[kshitijk4poor]

Thanks @UltraInstinct0x — the kanban-worker angle (spawn_failed × N, tasks going blocked) is a useful addition to the cron-side reports; the same fail-closed raise feeds both.

This landed on main about an hour after you opened, via #104152 (611ee856c5, authored by @gkd2323c, whose #102357 was the earliest of six PRs on this bug, opened 2026-09-03). Identical production change — OOMPolicy=kill removed from the shared _systemd_scope_argv — plus the same test flip you made (spawn argv asserts no OOMPolicy=) and a matching pin on the captured probe argv. Your test_scope_argv_carries_no_exec_only_properties overlaps those. Closing as superseded; #102486 and #103693 closed with the merge. If kanban still shows spawn_failed on a current main gateway, please open a fresh issue with the scope probe stderr — that would be a different rejection than this one.