`/api/messaging/platforms` still read a not-running gateway's retained
`gateway_state == "startup_failed"` verbatim, so a profile the operator
stopped with `hermes gateway stop` (which keeps the last failure and its
`exit_reason` on disk for diagnostics under `desired_state: stopped`) showed
every configured platform as `startup_failed` with the stale reason ("Port
8642 already in use") — the same symptom #112517 describes, on the sibling
surface `/api/status` had just been fixed for. The two docstrings promise
the sidebar strip and the Channels page "can never disagree on one page
load"; they did.
Lift the verdict into one shared `gateway.status.retained_gateway_state()`
used by both routers: `startup_failed` only while `desired_state` still
wants the gateway running, else `stopped`. Behaviour of `/api/status` is
unchanged (same condition, one call site); messaging now yields
`gateway_stopped` with no error for the operator-stopped case and keeps
`startup_failed` + `exit_reason` when the profile is meant to be running.
Invariant test `test_operator_stopped_gateway_does_not_report_retained_startup_failure`
(red on the previous head: state='startup_failed',
error_message='Port 8642 already in use').
Fixes#112517
595d4cfe45