The five #105417 tests collapse into one positive (fleet current at expected_sha
discharges the marker) and one parametrized negative (stale row / empty probe /
unknown identity / checkout moved past the marker keep the warning). Both are red
on origin/main. collect_fleet_versions' docstring now names why a live PID alone
is not a `current` row (#110420): write_runtime_status re-stamps pid/code_sha for
whatever process writes it, so the state-file fallback must pass
live_gateway_pid_for_home like the inventory already does (#109680).
teknium1
2fcb50c50e