DO NOT MERGE — regression replay to prove the journey gate goes red (#2336 AC 3)

[dolho]

Throwaway. Reverts ecf1327f (the #2092 fix) on top of dev so the 2026-08-14 "stopped agents will not start" regression is live again, and carries #2335/#2336's tier + gate.

Expected: journey-smoke FAILS on test_a_stopped_agent_can_be_started_again. If it passes, the gate does not have teeth and #2403 is not ready to be a required check.

Closing and deleting once observed.

[dolho]

Verified — this is what it existed for.

journey-smoke went RED on test_a_stopped_agent_can_be_started_again:

AssertionError: starting stopped agent 'pytest-ephemeral-journey-e15347a9'
answered 500: {"detail":"Failed to start agent (ValueError — details in backend logs)"}
— this is the 2026-08-14 regression: five of six stopped agents answered 500 here
and it reached users

Two things this replay established that a green run never would have:

1. The issue's replay instruction is imprecise. ecf1327f (bug: recreate_container_with_updated_config unconditionally starts the container — running-state precondition is caller-enforced and undocumented #2092) is the commit that CAUSED the regression; reverting it removes the bug. The fix to revert is 35a24d0c (bug: starting a stopped agent 500s when a recreate is needed — require_running guard not satisfied by start_agent_internal #2186). My first attempt reverted ecf1327f as written and the gate passed — correctly, because the code was fine.
2. The first version of the journey had no teeth. bug: starting a stopped agent 500s when a recreate is needed — require_running guard not satisfied by start_agent_internal #2186's title says the condition exactly: the 500 happened when a recreate is needed. A fresh agent stopped and started immediately has no config drift, so the bug is unreachable. The journey now changes resource limits while stopped to force the recreate path. Without that step it passed against the broken code — measured, not assumed.

Closing; branch deleted. Real work is #2403.