$ --status can remain executing while oomKilled=true for detached docker sessions

[konard]

$ --status can remain executing while oomKilled=true for detached docker sessions

Summary

Downstream link-assistant/hive-mind issue
link-assistant/hive-mind#2015 found two detached
Docker sessions where $ --status exposed oomKilled true but still reported
the primary status as executing.

Hive Mind can work around this by treating oomKilled true as terminal, but
the status contract would be safer if start-command made the session terminal
when Docker reports the container was OOM-killed.

Related upstream work

This is related to, but narrower than, closed issue
#144. Issue 144 requested
surfacing Docker resource-exhaustion markers. The remaining gap is that the
marker is visible while the top-level status can still remain executing.

Observed examples

Example 1 from downstream issue #2015:

sessionId 58ab247d-595e-4c7a-a2b5-0dffe839a2e7
internal session 1e9e7513-edd7-43a2-b143-169cfd794af6
status executing
oomKilled true
isolated docker

Example 2:

sessionId 526880a3-a1fc-45cf-8880-87d0a0d913f2
internal session d90880d4-aa05-4145-ac02-7542eea2041a
status executing
oomKilled true
isolated docker

The preserved start-command logs for both sessions did not contain normal
Finished / Exit Code footers, so downstream consumers only had the status
payload and Docker OOM marker to classify the result.

Expected behavior

When Docker reports State.OOMKilled=true, $ --status should return a
terminal session state. The exact status label can follow start-command naming,
but downstream tools need these invariants:

• top-level status is no longer executing;
• exit code is Docker's real exit code when known, otherwise 137 is a reasonable
  OOM/SIGKILL fallback;
• endTime or equivalent completion metadata is set when available;
• oomKilled true remains present in structured and links-notation output.

Downstream impact

Without this, consumers have to special-case oomKilled true ahead of the
primary status. Hive Mind's Telegram monitor had two bad outcomes:

• one OOM-killed session was reported with confusing killed copy while status
  still said executing;
• another OOM-killed session stayed visually stuck as executing.

Hive Mind PR link-assistant/hive-mind#2016 adds the
downstream workaround and regression tests.

References

• Downstream issue: Multiple tasks killed or stuck or false negatived at the same time (we need to fix all false positives, false negatives and errors) link-assistant/hive-mind#2015
• Downstream PR: Fix Docker OOM reporting and queue startup pacing link-assistant/hive-mind#2016
• Related upstream issue: Detached docker: surface container resource-exhaustion (OOM/disk) so a degraded exit-0 run isn't reported as clean success #144
• Docker Engine API reference for State.OOMKilled:
  https://docs.docker.com/reference/api/engine/version/v1.45/