fix(notifications): keep a known incident across a failed status fetch (#406)

The official CLI reports a status page it cannot reach as indicator
`unknown` (verified with CodexBar 0.67.0 in a network-isolated sandbox:
"Could not resolve host: status.claude.com"). The status memo stored that
severity over the known incident, keeping only its ID, so the next
successful fetch of the same incident ranked above `unknown` and was
announced again as a worsening.

`statusDecision` now keeps an existing baseline when the current severity
is `unknown`, without notifying. A replacement incident afterwards still
notifies. A first `unknown` with no incident baseline is unchanged.

Documentation: usage guide and the development notification contract
describe the preserved baseline.

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
930d8699d4Simone committed on 9/28/2026, 6:37:59 PM· committed by GitHubparent3925997
5 files changedLine totals unavailable