maintainer: kill-switch propagation into the manage/loop subsystem

[JSONbored]

The existing self-host kill-switch / dry-run control (already live for the maintainer auto-maintain layer, gittensory_get_automation_state) must extend to cover the NEW miner manage/loop subsystem: flipping the kill-switch must immediately halt any pending submission-gate check, any loop-reentry, and any in-flight CI/gate poll from proceeding to a write. Grouped with the other trigger-surface issues because a kill-switch that does not actually reach every write-adjacent code path is a safety-governor gap, not ordinary feature work.

Deliverables

• The manage-phase submission-gate and loop-reentry functions (from the sibling maintainer issues) both check the kill-switch/dry-run state as their FIRST guard, before any other logic
• A single shared helper (not duplicated per call site) that both consult, so future write paths can't accidentally skip the check
• Read-only polling (CI/gate watchers) is explicitly EXEMPTED from the kill-switch (observation continues; only actuation halts) — documented and tested
• Unit tests: kill-switch active blocks submission-gate and loop-reentry unconditionally regardless of otherwise-passing checks; kill-switch active does NOT stop the read-only pollers
• Manual owner review given this touches the safety-governor chokepoint

References

• src/mcp/server.ts:1439 (gittensory_get_automation_state) — the existing kill-switch/dry-run read surface this must be consulted through
• packages/gittensory-miner/src/manage/submission-gate.ts (new, from the submission-gate issue) — one call site
• packages/gittensory-miner/src/manage/loop-reentry.ts (new, from the loop-reentry issue) — the other call site
• git history: feat(agent): kill-switch + dry-run + action audit (#776) — the existing kill-switch this must reach into the new subsystem

[linear-code]

JSO-2436